1 PAISAGENS ACÚSTICAS E O USO DE MONITORAMENTO ACÚSTICO PASSIVO EM BIOMAS BRASILEIROS Eliziane Garcia de Oliveira Orientadora: Prof. Renata Santoro de Sousa-Lima Tese apresentada ao Programa de Pós Graduação em Ecologia da Universidade Federal do Rio Grande do Norte, como parte do requerimento para obtenção do título de Doutora em Ecologia. Natal, Fevereiro/2020 2 3 AGRADECIMENTOS Esse trabalho foi primeiramente fruto da confiança depositada em mim pela minha orientadora, Renata Sousa-Lima, a quem sou muito grata! Desde minha decisão de me juntar ao laboratório, de ser sua primeira aluna junto ao PPG em Ecologia, ela foi uma parte fundamental dessa jornada. Esses foram anos de muito aprendizado, e certamente tenho muito a aprender ainda. Que essa parceria ainda renda muitos frutos! Essa pesquisa não teria sido desenvolvida sem o auxílio fundamental do professor Miltinho Ribeiro. Além de ser uma inspiração pessoal e professional, desde meus tempos em Rio Claro, sua colaboração foi essencial antes e durante a realização desse projeto. Sou muito grata ao meu coorientador durante o período sanduíche, Paul Roe. Paul me recebeu em seu laboratório e sempre fez o possível para que eu me sentisse acolhida. Esses meses de estadia foram uma experiência muito enriquecedora, que não teria sido possível sem o auxílio dele. Eugenia e Juanka, que foram parceiros fundamentais na logística de campo, e me apresentaram à esse pedaço de mundo maravilhoso que é a Caatinga. Obrigada pelas refeições compartilhadas na carroceria da caminhonete, pela alegria e trabalho duro das noites com os morcegos. Seu João, essa figura incrível, por todo o carinho que tem por todos que passam pela Serra do Feiticeiro, e pela sua inestimável ajuda em campo. Jorge e Lara, pela ajuda em campo, perdidos por horas em meio a um mar de gravatás! Gabriel, sempre atento às plantinhas e sempre companheiro nos campos e na vida. Pela leitura e revisão cuidadosa de partes da tese, sou grata a Julia Verba, Manu Ramalho e Leo Gollo. Pela ajuda com meus desesperos estatisticos (cortes de cabelo e boas risadas), Marina Fagundes. Carlos Gussoni, pelas identificações de espécies de aves rápidas e certeiras. Também dedicaram um tempo a me ajudar com análises da tese, Camila Teixeira, Gustavo Paterno, Giba. Muito obrigada! Ao pessoal do PPG, professores e alunos, pelos dias de conversa em meio a trabalho no departamento, pelas festinhas cheias de samba. Não ouso escrever aqui o nome de todos, na certeza de que vou esquecer muita gente. 4 Minha estadia em Brisbane não teria sido a mesma sem a companhia de pessoas muito queridas e muito papo na beira do rio. Ju, Le, Alanny, Natis, Marina, Daniel, os “manorists” Jesses e Albie. Celeste, por sempre nos receber com um sorriso. Ao pessoal do laboratório na QUT, tão diversos e curiosos: Jess, Marina, Tshering, Anthony, Phill, Michael, David, Jinglan, Manoosh, Miriem, Kelly. Ao pessoal do LaB, um time dedicado e apaixonado pelo que faz. Lara, Heloise, Alan e Jorge tiveram um papel muito importante no desenvolvimento desse trabalho. Luane, que além do convívio no LaB, dividiu comigo sempre de bom humor uma viagem pela tão diferente Índia. Essa jornada foi marcada por muitas mudanças desde o começo. E eu sempre me senti muito grata por todas as pessoas que fizeram parte desses anos. Laurinha, alimentou em mim essa ideia bonita de vir morar na beira do mar, me deu todo o apoio pra que isso acontecesse (e depois foi embora, mas tudo bem). À Julia e Edu, por serem minha primeira casa nessa cidade que me acolheu, me chacoalhou e me ensinou tanto. Minhas primeiras lembranças de Natal são dos nossos cafés, que viravam almoço, que viravam janta, na varanda da Pinga. Jota, que mesmo na correria estava sempre animando a casa e os rolês. Às verms, essas mulheres incríveis e tão inspiradoras. Helo, por dividir a casa, a campervan, paisagens incríveis do outro lado mundo, muitas risadas e uma dúzia de de bichos comigo. Eugenia, pelos momentos de total fuga da realidade, e por me trazer de volta ao chão. Julia, que me acolheu também do outro lado do mundo. Ju, nascemos pra isso, ainda moraremos no mar! Minhas lembranças com vocês não tem preço! Ao pessoal da Vila Feliz, esse não-condomínio que me acolheu e que será sempre meu lar. Nani, Daniel e Ju… esses anos não teriam sido os mesmos sem vocês! Obrigada por todas as praias de final de semana com Cação e dindin, cafés, almoços, jantas, tretas. Tive o prazer de conviver com muita gente maravilhosa nesse lugar tão especial, dividindo aulas de yoga e muita comida, sempre. Tenho também uma gratidão sem tamanho para com os bichinhos que passaram pela minha vida. Piper, minha companheira de aventuras, as terríveis tigradinhas Mangaba e Cajuína e a sempre tranquila Ritinha. 5 Ao pessoal do Climbar, companheiros de escalada, que me deu uma outra visão da Caatinga. E me ensinou que na pedra e na vida, a gente perde as forças se fica em uma agarra por medo de seguir. Toca pra cima! Sempre presentes, minha grande família do coração em Rio Claro. Jesus, uma amizade pra vida toda! À (parte da) minha turma de graduação que ainda se encontraria todas as quartas se estivesse na mesma cidade, mas continua dividindo as dores e delícias da vida. Van, Pam, Manu, Monão, Bruna, obrigada por todos esses anos de convivência! O pessoal do laboratório de herpetologia, com quem dividi incontáveis festas juninas. Ao Leo, que fez do nosso universo paralelo uma caminhada muito real. É bom partilhar a vida boa com você! À minha família que sempre me apoiou na decisão de voar pra bem longe de casa, mesmo em momentos difíceis. O presente trabalho foi realizado com apoio da Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Código de Financiamento 001 (This study was financed by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001). 6 7 RESUMO O estudo de paisagens acústicas vem ganhando atenção da comunidade científica nos últimos anos, mostrando-se como uma importante ferramenta na avaliação tanto de saúde ambiental como dos efeitos de mudança no uso da terra e mudanças no clima em várias escalas temporais e espaciais. Em ambientes tropicais, a alta biodiversidade, associada ao conhecimento taxonômico mais restrito para muitos grupos (como insetos) e menor investimento financeiro em pesquisas, faz com que estudos de ecologia acústica sejam mais desafiadores que em outras regiões. Por serem também as áreas de grande importância para conservação, o uso de monitoramento acústico mostra-se ainda mais relevante na medida em que pode facilitar estudos de longo prazo a um custo mais baixo que tradicionais estudos ecológicos. Diante de um panorama de constantes inovações tecnológicas e métodos de análise de dados na área de ecologia acústica, essa tese teve como objetivo geral avaliar o uso do monitoramento acústico passivo em biomas brasileiros, fornecendo protocolos de análise e contribuindo com o avanço da ecologia acústica em áreas tropicais. A tese está dividida em três capítulos e um apêndice com outras publicações não relacionadas diretamente à tese. No primeiro capítulo descrevemos padrões de utilização do espaço acústico em uma área de Caatinga, através do agrupamento em clusters de múltiplos índices acústicos. Na estação seca, o som do vento foi predominante nas gravações, enquanto na estação chuvosa, observamos alta presença de biofonia. Variações diárias incluíram aves e vento ocupando o espaço acústico durante o dia, enquanto insetos predominaram no período noturno. No segundo capítulo, também em uma área de Caatinga, nós buscamos compreender como um parque eólico modifica a paisagem acústica ao seu redor. Encontramos que a atividade biofônica aumenta com a distância dos aerogeradores, sugerindo que aves e insetos tem seus padrões de atividade acústica modificados em resposta à presença das turbinas. No terceiro capítulo utilizamos dados de Caatinga, Cerrado e Mata Atlântica, para testar a eficiência de índices acústicos em predizer a riqueza de espécies de aves e a discriminação de composição de comunidades entre os biomas. A performance dos índices acústicos variou entre os habitats, e características das comundiades de aves como predominância de espécies e faixas de frequência utilizadas por elas podem estar afetando os resultados dos índices. Mesmo assim, uma combinação de índices acústicos foi capaz de diferenciar as comunidades dos três biomas. Palavras-chave: Índices Acústicos; Paisagens Acústicas; Ecologia Acústica; Floresta sazonalmente seca; Caatinga; Cerrado; Mata Atlântica 8 ABSTRACT Soundscape studies has becoming more popular nowadays as an important tool with the potential to evaluate environmental health, changes in land use, climate change, in several temporal and spatial scales. In tropical environments, the high biodiversity associated with a restricted taxonomic knowledge for many taxa and the lack of financial resources make the acoustic ecology studies more challenging than in other regions. At the same time, tropical areas are important in terms of biodiversity conservation, and the use of acoustic monitoring is even more relevant once it can facilitate long-term studies at a lower cost than traditional ecological surveys. Facing a reality of rapid technological innovations and data analysis methods in the acoustic ecology field, this thesis has the main goal of evaluate the use of passive acoustic monitoring in Brazilian biomes, providing analysis protocols and contributing to advance of acoustic ecology in tropical regions. The thesis is divided in three chapters and an appendix with other production non-related to the thesis. In the first chapter we describe patterns in use of acoustic space in a Caatinga area by clustering multiple acoustic indices. During dry season, wind was predominant in the recordings, while in rainy season, was biophony. Daily variations included the birds and wind occupying the acoustic space during the day, while insects predominated at night. In the second chapter, also in a Caatinga area, we aimed to understand how a wind farm facility modifies the soundscape around it. We’ve found that biophonic activity decreases as we move closer to the turbines, suggesting that both birds and insects have their acoustic activity patterns modified in response to wind turbines presence. In the third chapter, we used data from Caatinga, Cerrado and Atlantic Forest, to test whether single indices are able to predict bird species richness and if a composition of several indices allow discrimination of different species’ composition. We’ve found that single indices performance is variable among habitats, and features of birds’ communities as predominance of species and frequency bands used may be affecting the indices results. Even so, a combination of indices was able to distinguish among biomes. Keywords: Acoustic Indices; Soundscape; Acoustic Ecology; Seasonally Dry Tropical Forest; Caatinga; Cerrado; Atlantic Forest 9 INTRODUÇÃO GERAL Essa seção tem como objetivo apresentar as bases teóricas que deram origem a esse trabalho. O texto apresenta uma explicação de conceitos e abordagens mais comuns utilizadas na área de ecologia acústica. Embora escrito em português, apresentamos a tradução dos termos-chave em inglês entre parêntesis, uma vez que é o idioma utilizado nos três capítulos. Para tanto, dividimos em quatro subseções: (1) Conceitos fundamentais no estudo de paisagens acústicas, (2) Uso de monitoramento acústico passivo em estudos ecológicos, (3) Conservação de ambientes acústicos saudáveis, (4) Abordagens metodológicas e desafios práticos. 1. Conceitos fundamentais no estudo de paisagens acústicas Os sons são utilizados por diversos grupos animais como principal forma de comunicação e, portanto, são foco de estudo de uma área de conhecimento conhecida como bioacústica. A bioacústica investiga a função dos sinais acústicos na comunicação animal, e é historicamente uma área mais voltada às interações entre indivíduos. O comitê técnico de Bioacústica Animal da Acoustical Society of America (https://tcabasa.org/) inclui que em escalas mais amplas, o som não é apenas uma troca de energia e informação, mas é capaz de mediar (e até mesmo simbolizar) relações entre emissores, receptores e seus ambientes, refletindo um sistema dinâmico de comportamentos característico de uma abordagem ecológica. O termo paisagem acústica (soundscape) foi proposto pela primeira vez em um contexto urbano, definido como propriedades acústicas das cidades que ajudam as pessoas a 10 se localizarem em certos espaços (Southworth, 1969). Ainda hoje, é um conceito utilizado no planejamento de cidades e manifestações artísticas (Raimbault & Dubois, 2005). A partir de 1977 começou a ser usado também para abordar questões relacionadas a ambientes naturais. Schafer (1977) define uma paisagem acústica como sendo “as características acústicas de uma área que refletem seus processos naturais” (uma retrospectiva histórica mais detalhada pode ser encontrada em Zorzetto, 2019). Os componentes de uma paisagem acústica (Figura 1) incluem emissões sonoras de origem biológica – biofonia (biophony), de origem em processos geofísicos – geofonia (geophony) e antropogênica – antropofonia (antropophony). Biofonia inclui todos os sons emitidos por animais, também chamados zoofonia (zoophony) (Ferreira et al., 2018). Geofonia inclui sons como chuva, vento, e trovões. Antropofonia engloba os sons advindos de atividades urbanas e industriais (Krause, 1987; Pijanowski et al., 2011). Esses três componentes ocupam o mesmo espaço acústico, e assim, fenômenos acústicos oriundos de geofonia ou antropofonia interagem, modificam e tem o potencial de impactar a biofonia. Figura 1 – Fontes de emissão sonora que compõe uma paisagem acústica. 11 Considerando essa interação entre os três componentes da paisagem acústica, três hipóteses diferentes e não mutuamente exclusivas foram propostas para explicar como as características vocais de diferentes espécies animais variam e se adaptam ao ambiente no qual evoluíram: Hipótese de Adaptação Morfológica, de Adaptação Acústica e de Nicho Acústico. A Hipótese de Adaptação Morfológica (Morphological Adaptation Hypothesis – MAH) refere-se às limitações morfológicas e fisiológicas de determinado organismo, que restringem as características das suas emissões acústicas (Wallschlager, 1980). As características vocais das espécies também são moldadas e selecionadas ao longo do tempo por fatores ambientais como cobertura vegetal, compactação do solo, meio de transmissão. Esse fenômeno é conhecido como Hipótese da Adaptação Acústica (Acoustic Adaptation Hypothesis – AAH) (Ey & Fischer, 2009; Morton, 1975). Essas duas hipóteses explicam em parte as pressões que levam as espécies a terem distintos padrões de emissão acústica. Uma terceira pressão seria a de minimizar sobreposições no tempo e espaço entre sinais sonoros, maximizando a transmissão de informação ao receptor. A essa hipótese damos o nome de Hipótese de Nicho Acústico (Acoustic Niche Hypothesis) (Krause, 1993). Segundo ela, cada espécie teria seu próprio nicho acústico no espectro de tempo-frequência e, ambientes mais antigos, mais diversos e menos perturbados possuiriam nichos acústicos amplamente ocupados. Modificações nessa ocupação seriam então possivelmente um sinal de perturbações ambientais (Pijanowski et al., 2011) ou alterações em processos ecológicos. 12 2. Uso de Monitoramento acústico passivo em estudos ecológicos O monitoramento acústico passivo (MAP) consiste na gravação de arquivos de áudio sem que haja a necessidade da presença do pesquisador operando o gravador. Esses sistemas de gravação automáticos permitem que amostragens sejam feitas em uma larga escala espaço-temporal, reduzindo os esforços e gastos com trabalhos de campo, além de diminuir potenciais impactos que os observadores possam ter sobre o comportamento natural dos animais. MAP envolve a gravação de sons utilizando uma variedade de sensores acústicos, a depender do estudo, e uma posterior extração de dados relevantes a partir dos áudios (Fig. 2). Até o momento, a maior parte dos trabalhos de MAP em ambientes terrestres tem um foco voltado para estudos com quirópteros, em regiões temperadas, com gravações feitas no período noturno e incluem uma análise manual dos arquivos (Sugai et al., 2019). Um produto das gravações de áudio é a amostragem de espécies que podem ser identificadas a partir de dados acústicos. Nesse sentido, Shonfield & Bayne (2017) avaliaram a eficácia desse método comparado a pontos de escuta (um método de amostragem tradicionalmente utilizado) chegando à conclusão de que, de um modo geral, a análise de dados provenientes de MAP é bastante eficiente e permite que diversos pontos sejam amostrados ao mesmo tempo e que os dados sejam armazenados e avaliados por mais de um especialista. Uma limitação dessa metodologia é para grupos de organismos que não apresentam suas emissões sonoras identificadas e catalogadas, como é o caso dos ortópteros (Riede, 2017). 13 Figura 2 – Workflow de estudos de monitoramento acústico passivo (modificado de Gibb et al., 2019) 3. Conservação de ambientes acústicos saudáveis As mudanças antropogênicas no ambiente acústico, especialmente desde a revolução industrial, vêm aumentando substancialmente, de forma que dificilmente é possível encontrar um local no mundo, seja na superfície da terra ou no fundo dos oceanos, que tenha suas características sonoras naturais totalmente preservadas. Os impactos acústicos do crescente aumento de ruído no ambiente são bem estudados em humanos e seus efeitos incluem alterações no ritmo de sono, aumento do risco de doenças cardíacas e prejuízos auditivos e cognitivos (Fritschi et al., 2011). O efeito da poluição sonora sobre outras espécies de animais é mais bem estudado em ambientes aquáticos, mas vêm ganhando atenção em ambientes terrestres mais recentemente (Shannon et al., 2015). De acordo com a literatura, a vida selvagem apresenta repostas ao ruído a níveis sonoros acima de 40dB, mas quantificar os efeitos de antropofonia em animais é um desafio, já que a sensibilidade auditiva varia de acordo com os taxa, contextos e histórias de vida (Shannon et al., 2015). Essas fontes de alteração da paisagem sonora também não podem ser desconectadas de outras mudanças, como alteração do habitat e distúrbio visual, o que 14 dificulta uma interpretação precisa das respostas biológicas ao ruído (Summers, Cunnington, & Fahrig, 2011). O uso de soundscapes em estudos relacionados a conservação vêm crescendo nos últimos anos, uma vez que a comparação das paisagens sonoras em áreas sob pressões antrópicas diferentes pode nos fornecer um entendimento rápido da intensidade dos impacto causados (Astaras, Linder, Wrege, Orume, & Macdonald, 2017; Burivalova et al., 2017) e fornece uma ferramenta eficiente e relativamente barata para monitorar tendências. Burivalova et al. (2019) aponta o potencial do uso de monitoramento acústico associado a imagens de satélite para avaliação do sucesso de medidas de conservação em áreas tropicais. O uso dessas ferramentas é também recomendado em sistemas de produção agrícola que, embora não naturais, terão um papel cada vez mais importante na manutenção da biodiversidade (Doohan et al., 2019). Além disso, gravações acústicas constituem um banco de dados extenso e que pode ser utilizado por um longo período de tempo, analisado sob diferentes óticas e múltiplos pesquisadores e, portanto, vêm sendo consideradas como “cápsulas de tempo”, e serão criticamente importantes para comparações nas próximas décadas (Sugai & Llusia, 2019). 4. Abordagens metodológicas e desafios práticos Uma vez que estudos de ecologia acústica possuem focos diferentes, encontramos na literatura uma ampla variação em equipamentos utilizados, frequências amostrais e esquemas temporais de amostragem. De acordo com Phillips (2018), a partir de uma revisão bibliográfica de trabalhos de paisagem acústica, a maioria dos trabalhos utiliza uma frequência de amostragem de 44.1 ou 48 kHz, o que significa que emissões sonoras de até 22 15 ou 24 kHz, respectivamente, serão amostradas com qualidade (é preciso um mínimo de duas amostras por ciclo – Teorema de Nyquist, Nyquist, 1928). O esquema temporal de amostragem pode variar de gravações contínuas (Phillips et al., 2018) a até alguns minutos por dia (Machado et al., 2017; Sugai et al., 2019). A exploração de resultados provenientes de diferentes esquemas de amostragem (Pieretti et al., 2015) são importantes nessa etapa de desenvolvimento da ecologia acústica e visam auxiliar inclusive na redução de gastos relacionados a alimentação de equipamentos de gravação, armazenamento de dados e subsequentes processamento e análise. Nos últimos anos observamos uma significativa redução de custos associados a equipamentos e tecnologia de armazenagem e análise. Enquanto há poucos anos apenas uma empresa fornecia a maior parte de gravadores autômatos utilizados em estudos de ecologia acústica, hoje open-source hardwares como Audiomoth (Hill et al., 2018) permitem que estudos de monitoramento acústico sejam feitos a um custo mais baixo. Essa facilidade permite também que grandes bancos de dados sejam criados e, consequentemente, novos desafios para a sua análise surjam. A análise de grandes bancos de dados acústicos não pode ser feita da mesma forma que estudos de bioacústica com foco em poucas espécies ou em restritos espaços de tempo. Para solucionar esse impasse no processamento dos dados, começaram a ser desenvolvidos índices acústicos, métricas que resumem determinadas características presentes nas gravações e que podem refletir processos ecológicos presentes nelas (Sueur et al., 2014). A avaliação da eficiência dessas métricas ainda é incipiente (Ferreira et al., 2018; Gasc et al., 2015; Gasc et al., 2013). A eficiência de certos índices em refletir diversidade de aves por exemplo, parece variar de acordo com a região (Eldridge et al., 2018). A utilização de índices para monitorar e diferenciar taxa distintos dentro das mesmas gravações representa um 16 desafio a mais no estudo de soundscapes, especialmente em áreas com alta biodiversidade (Ferreira et al., 2018). A combinação de índices acústicos está sendo usada mais recentemente para auxiliar nessa complexa caracterização das paisagens acústicas (Phillips et al., 2018). Índices acústicos também são utilizados para auxiliar na visualização de grandes bancos de dados. Em trabalhos de bioacústica, utiliza-se espectrogramas, representações visuais do som em três dimensões: frequência, tempo e amplitude. Em gravações de longa duração, entretanto, a utilização de simples espectrogramas torna-se inviável. Índices acústicos associados à visualização em espectrogramas de falsa cor (False Colour Spectrograms - FCS) têm fornecido resultados interessantes em grandes bancos de dados e facilitado a interpretação (Towsey et al., 2014). Os FCS estão sendo usados para identificar padrões em áudios de longa duração e, mais recentemente, pesquisadores vêm analisando sua eficácia para identificação de grupos taxonômicos ou espécies (Towsey et al., 2018). OBJETIVO GERAL E ORGANIZAÇÃO DA TESE Diante desse panorama de constantes inovações tecnológicas e métodos de análise de dados, essa tese teve como objetivo geral avaliar o uso do monitoramento acústico passivo em biomas brasileiros, fornecendo protocolos de análise e contribuindo com o avanço da ecologia acústica em áreas tropicais. A tese está dividida em três capítulos e contém três apêndices: um texto de divulgação científica publicado na revista Ciência Hoje Crianças, e duas publicações não relacionadas diretamente com a tese. No primeiro capítulo descrevemos padrões de utilização do espaço acústico em uma área de Caatinga, analisando variações sazonais e padrões diários da biofonia. No segundo capítulo, também em uma área de Caatinga, buscamos responder como 17 um parque eólico modifica a paisagem acústica do seu entorno. No terceiro capítulo utilizamos dados de Caatinga, Cerrado e Mata Atlântica para responder como índices acústicos refletem diferenças em riqueza e composição de comunidade de aves nos três biomas. Referências Bibliográficas Astaras, C., Linder, J. M., Wrege, P., Orume, R. D., & Macdonald, D. W. (2017). Passive acoustic monitoring as a law enforcement tool for Afrotropical rainforests. Frontiers in Ecology and the Environment, 15(5), 233–234. https://doi.org/10.1002/fee.1495 Burivalova, Z., Game, E. T., & Butler, R. A. (2019). The sound of a tropical forest. Science, 363(6422), 28–29. https://doi.org/10.1126/science.aav1902 Burivalova, Z., Towsey, M., Boucher, T., Truskinger, A., Apelis, C., Roe, P., & Game, E. T. (2017). Using soundscapes to detect variable degrees of human influence on tropical forests in Papua New Guinea. Conservation Biology, 1–29. https://doi.org/10.1111/cobi.12968 Doohan, B., Fuller, S., Parsons, S., & Peterson, E. E. (2019). The sound of management: Acoustic monitoring for agricultural industries. Ecological Indicators, 96(September 2018), 739–746. https://doi.org/10.1016/j.ecolind.2018.09.029 Eldridge, A., Guyot, P., Moscoso, P., Johnston, A., Eyre-Walker, Y., & Peck, M. (2018). Sounding out ecoacoustic metrics: Avian species richness is predicted by acoustic indices in temperate but not tropical habitats. Ecological Indicators, 95(December), 939–952. https://doi.org/10.1016/j.ecolind.2018.06.012 Ey, E., & Fischer, J. (2009). The “acoustic adaptation hypothesis”—a review of the evidence from birds, anurans and mammals. Bioacoustics, 19(1–2), 21–48. https://doi.org/10.1080/09524622.2009.9753613 Ferreira, L. M., Oliveira, E. G., Lopes, L. C., Brito, M. R., Baumgarten, J., Rodrigues, F. H., & Sousa-Lima, R. S. (2018). What do insects, anurans, birds, and mammals have to say about soundscape indices in a tropical savanna. Journal of Ecoacoustics, 2(March), PVH6YZ. https://doi.org/10.22261/JEA.PVH6YZ Fritschi, L., Brown, A. L., Kim, R., Schwela, D. H., & Kephalopoulos, S. (2011). Burden of Disease from Environmental Noise – Quantification of Healthy Life Years Lost in Europe. Retrieved from http://www.who.int/ quantifying_ehimpacts/publications/e94888/en/ Gasc, A., Pavoine, S., Lellouch, L., Grandcolas, P., & Sueur, J. (2015). Acoustic indices for biodiversity assessments: Analyses of bias based on simulated bird assemblages and recommendations for field surveys. Biological Conservation, 191(January), 306–312. https://doi.org/10.1016/j.biocon.2015.06.018 18 Gasc, Amandine, Sueur, J., Pavoine, S., Pellens, R., & Grandcolas, P. (2013). Biodiversity Sampling Using a Global Acoustic Approach: Contrasting Sites with Microendemics in New Caledonia. PLoS ONE, 8(5). https://doi.org/10.1371/journal.pone.0065311 Gibb, R., Browning, E., Glover-Kapfer, P., & Jones, K. E. (2019). Emerging opportunities and challenges for passive acoustics in ecological assessment and monitoring. Methods in Ecology and Evolution, 10(2), 169–185. https://doi.org/10.1111/2041-210X.13101 Hill, A. P., Prince, P., Piña Covarrubias, E., Doncaster, C. P., Snaddon, J. L., & Rogers, A. (2018). AudioMoth: Evaluation of a smart open acoustic device for monitoring biodiversity and the environment. Methods in Ecology and Evolution, 9(5), 1199–1211. https://doi.org/10.1111/2041-210X.12955 Krause, B. (1987). The Niche Hypothesis: How animals taught us to Dance and Sing, 1– 6. Krause, B. L. (1993). The Niche Hypothesis: A Virtual Symphony of Animal Sounds, The Origins of Musical Expression and the Health of Habitats. The Soundscape Newsletter, (6), 6– 10. Retrieved from http://interact.uoregon.edu/Medialit/wfae/library/newsletter/SNL6.PDF Machado, R. B., Aguiar, L., & Jones, G. (2017). Do acoustic indices reflect the characteristics of bird communities in the savannas of Central Brazil? Landscape and Urban Planning, 162, 36–43. https://doi.org/10.1016/j.landurbplan.2017.01.014 Morton, E. S. (1975). Ecological Sources of Selection on Avian Sounds. The American Naturalist, 109(965), 17–34. https://doi.org/10.1086/282971 Nyquist, H. (1928). Certain Topics in Telegraph Transmission Theory. Transactions of the American Institute of Electrical Engineers, 47(2), 617–644. https://doi.org/10.1109/T- AIEE.1928.5055024 Phillips, Y. (2018). THESIS: Analysis and Visualisation of very-long-duration acoustic recordings of the natural environment. Queensland University of Technology, Brisbane. Phillips, Y. F., Towsey, M., & Roe, P. (2018). Revealing the ecological content of long- duration audio-recordings of the environment through clustering and visualisation. PLoS ONE, 13(3), 1–27. https://doi.org/10.1371/journal.pone.0193345 Pieretti, N., Duarte, M. H. L., Sousa-Lima, R. ., Rodrigues, M., Young, R. J., & Farina, A. (2015). Determining temporal sampling schemes for passive acoustic studies in different tropical ecosystems. Tropical Conservation Science, 8(1), 215–234. Pijanowski, B. C., Villanueva-rivera, L. J., Dumyahn, S. L., Farina, A., Krause, B. L., Napoletano, B. M., … Pieretti, N. (2011). Soundscape Ecology : The Science of Sound in the Landscape. BioScience, 61(3), 203–216. https://doi.org/10.1525/bio.2011.61.3.6 Raimbault, M., & Dubois, D. (2005). Urban soundscapes: Experiences and knowledge. Cities, 22(5), 339–350. https://doi.org/10.1016/j.cities.2005.05.003 Riede, K. (2017). Acoustic profiling of Orthoptera for species monitoring and discovery: present state and future needs. PeerJ Preprints, (November). https://doi.org/10.7287/peerj.preprints.3397v1 Schafer, R. M. (1977). The soundscape: our sonic environment and the tuning of the world. Rochester, Vt.: Destiny Books. 19 Shannon, G., Mckenna, M. F., Angeloni, L. M., Crooks, K. R., Fristrup, K. M., Brown, E., Wittemyer, G. (2015). A synthesis of two decades of research documenting the effects of noise on wildlife. Biological Reviews, 24. https://doi.org/10.1111/brv.12207 Shonfield, J., & Bayne, E. M. (2017). Autonomous recording units in avian ecological research: current use and future applications. Avian Conservation and Ecology, 12(1), art14. https://doi.org/10.5751/ACE-00974-120114 Southworth, M. (1969). The Sonic Environment of Cities. Environment and Behavior, 50–70. Retrieved from file:///Users/mohdsayeed/Desktop/laptu desktop/others/etc etc/SAGE JOURNALS/SAGE JOURNALS/environment and behavior/june 1969/the sonic environment of cities.pdf Sueur, J., Farina, A., Gasc, A., Pieretti, N., & Pavoine, S. (2014). Acoustic indices for biodiversity assessment and landscape investigation. Acta Acustica United with Acustica, 100(4), 772–781. https://doi.org/10.3813/AAA.918757 Sugai, L. S., Llusia, D., Sanna, T., Silva, F., & Desjonqueres, C. (2019). A roadmap for survey designs in terrestrial acoustic monitoring, Remote Sensing in Ecology and Conservation, 1–16. https://doi.org/10.1002/rse2.131 Sugai, L. S. M., & Llusia, D. (2019). Bioacoustic time capsules: Using acoustic monitoring to document biodiversity. Ecological Indicators, 99(January), 149–152. https://doi.org/10.1016/j.ecolind.2018.12.021 Sugai, L. S. M., Silva, T. S. F., Ribeiro, J. W., & Llusia, D. (2019). Terrestrial Passive Acoustic Monitoring: Review and Perspectives. BioScience, 69(1), 5–11. https://doi.org/10.1093/biosci/biy147 Summers, P. D., Cunnington, G. M., & Fahrig, L. (2011). Are the negative effects of roads on breeding birds caused by traffic noise? Journal of Applied Ecology, 48(6), 1527–1534. https://doi.org/10.1111/j.1365-2664.2011.02041.x Towsey, M., Zhang, L., Cottman-Fields, M., Wimmer, J., Zhang, J., & Roe, P. (2014). Visualization of long-duration acoustic recordings of the environment. Procedia Computer Science, 29, 703–712. https://doi.org/10.1016/j.procs.2014.05.063 Towsey, M., Znidersic, E., Broken-Brow, J., Indraswari, K., Watson, D. M., Phillips, Y., … Roe, P. (2018). Long-duration, false-colour spectrograms for detecting species in large audio data-sets. Journal of Ecoacoustics, 2, IUSWUI. https://doi.org/10.22261/JEA.IUSWUI Truax, B., & Barrett, G. W. (2011). Soundscape in a context of acoustic and landscape ecology. Landscape Ecology, 26(9), 1201–1207. https://doi.org/10.1007/s10980-011-9644-9 Wallschlager, D. (1980). Correlation of song frequency and body weight in passerine birds. Experientia, 36, 412. Zorzetto, R. (2019). A acústica do ambiente. Revista FAPESP, 281, 64–67. 20 21 The Caatinga Orchestra: Clustering Multiple Acoustic Indices track Temporal Changes in a Seasonally Dry Tropical Forest Eliziane Garcia de Oliveiraa,b,d*, Milton Cezar Ribeiroc, Paul Roeb, Renata S. Sousa-Limaa,d a Laboratório de Bioacústica (LaB), Departamento de Fisiologia e Comportamento, Biosciences Center, Universidade Federal do Rio Grande do Norte, Avenida Senador Salgado Filho, 3000, Bairro Lagoa Nova, Natal, RN. 59078-970, Brazil. b Science and Engineering Faculty, Queensland University of Technology. 2 George St, Brisbane City QLD 4000, Australia. c Spatial Ecology and Conservation lab (LEEC), Department of Ecology, State University of Sao Paulo (UNESP), Av. 24 A, 1515, Rio Claro SP, 13506-900, Brazil. d Graduate Program in Ecology, Biosciences Center, Universidade Federal do Rio Grande do Norte, Avenida Senador Salgado Filho, 3000, Bairro Lagoa Nova, Natal, RN. 59078-970, Brazil. *Corresponding author: eliziane.garcia@gmail.com Submitted to: Ecological Indicators Highlights  Clustering of eleven acoustic indices were used to discriminate the main sound sources present in the Caatinga soundscape  The diurnal period was dominated by sounds of birds and wind, while at night insect acoustic activity was prevalent  Acoustic indices revealed seasonal differences in animal activity  Biophony (insects and birds) was more prevalent during the rainy season while in the dry season, when vegetation is deprived of leaves, wind (geophony) dominated  Entropy of the Spectral Peaks (EPS) was influenced by birds and orthopterans. Orthopteran activity also drove the diel patterns of Entropy of Average Spectrum (EAS) and High Frequency Cover (HFC) indices. 22 23 Abstract Advances in technologies for data acquisition, storage and analysis have boosted Acoustic Ecology studies, but there is still a lack of protocols and it is unknown which methodologies can be applied to answer ecological questions in different environments with varying temporal and spatial dynamics. Tropical forests are generally more complex than temperate ones, in terms of use of acoustic space and species diversity. The seasonally dry tropical forest in Brazil is a threatened biome, with two marked seasons that shape the vegetation and animal activity patterns. In this study, we investigate the applicability of passive acoustics in monitoring seasonally dry tropical forests (SDTF), describing the soundscape and tracking diel patterns and seasonal changes. Combining multiple indices, visualization through false colour spectrograms and clustering we describe the acoustic activity of the main faunal groups that compose the biophonic orchestra in a SDTF area in Northeast Brazil. Distinct patterns were found between day – when birds and wind were the main sound sources – and night – with Orthopterans occupying a large frequency band. Other sound sources in the SDTF soundscape included cicada, rain, and anthropogenic interference such as domestic animals, cars and gunshots. Clustering of eleven acoustic indices was useful to distinguish sound patterns from several sources, especially in the dry season. Further investigation within each cluster showed specific relationships among selected indices and different sound sources. Birds were associated with Entropy of the Spectral Peaks (EPS) and Orthopteran also had a relationship with EPS, as well as with Entropy of Average Spectrum (EAS) and High Frequency Cover (HFC). Variation in diel values of these selected indices, as well as the number of samples included in each cluster category, were successfully used to describe the acoustic activity of Birds and Orthopteran and to track changes between rainy and dry seasons. A better understanding of the soundscape dynamics in a highly seasonal tropical environment was achieved by applying cheap and reliable novel methodologies to study biodiversity in geopolitical regions where funding for conservation initiatives is limited. Keywords: Acoustic Indices, Soundscape, Biophony, Geophony, Acoustic Ecology 24 1. Introduction The use of acoustic data to study ecological processes relies on the fact that sounds play a fundamental role for many species and ecosystems, being used by vertebrates and invertebrates in a variety of functions and contexts (Bradbury & Vehrencamp, 1998). Employment of acoustic tools is non-invasive and provides data about variation on temporal and spatial patterns of species richness (Depraetere et al., 2012; Duarte et al., 2015) and changes in behaviour (Parks et al., 2011). Passive acoustic monitoring (PAM) has become popular due to recent advances in technologies for data acquisition, storage and analyses (Burivalova et al., 2019; Gibb et al., 2019; Merchant et al., 2015). Major caveats to these methods include that they only work for vocal species and it is difficult to identify individuals calling animals for most taxa (Gibb et al., 2019). The popularization of autonomous passive acoustic monitoring was followed by the exponential growth of soundscape ecology (Pijanowski et al., 2011). The term “soundscape” appeared for the first time in an urban context (Southworth, 1969) and describes the acoustic characteristics of a given area that reflects their processes (Schafer, 1977) and, in natural environments, creates dynamic acoustic patterns (Pijanowski et al., 2011). We can recognize three different sound sources in a soundscape (Krause, 1987): biophony, arguably zoophony (Ferreira et al. 2018) (sounds produced by animals, usually related to communication), geophony (sounds from physical processes, like wind, rain, thunder), and anthropophony (sounds from human made activities or machines). The study of soundscapes has applications in several temporal and spatial scales to address questions about environmental health, changes in land use, and climate change, to name a few (Blumstein et al., 2011; Burivalova et 25 al., 2019; Buxton et al., 2018; Farina, 2014; Krause & Farina, 2016; Pijanowski et al., 2011; Tucker et al., 2014). As a relatively new research field, with a potential to generate an enormous amount of data, researchers are now dealing with data management and wish to find fast and reliable ways to visualize and interpret acoustic big data (Phillips et al., 2018; Sueur et al., 2014; Towsey et al., 2014). The use of acoustic indices to summarize the information in audio-recordings has been explored. An acoustic index is an statistic that summarizes some aspects of structure and distribution of acoustic energy, and potentially reflects ecological processes (Towsey et al., 2014). The relationship between acoustic indices and community diversity are still poorly understood (Burivalova et al., 2019; Ferreira et al., 2018; Gibb et al., 2018) and, besides the great potential to be used in a more broad context (Aide et al., 2017), most literature still focus on avian community (Machado et al., 2017; Pieretti et al., 2015; Zhang et al., 2016). Given the diversity of acoustic patterns within a recording, it is not expected that a single index could precisely describe the soundscape (Buxton et al., 2018; Ferreira et al., 2018). Therefore, recent studies are proposing the combination of several non-redundant indices to achieve best results (Phillips et al., 2018). More visual approaches in characterizing soundscapes rely on the idea that is impossible to identify patterns in long duration recordings by only listening to them (Sankupellay et al., 2015). To address this caveat, multiple acoustic indices’ visualization has been proposed using two techniques: false colour spectrograms (Sankupellay et al., 2015) and data ordination (Phillips et al., 2018; Sankupellay et al., 2015). Here we propose the utilization of multiple visualization tools to validate the applicability of passive acoustic monitoring in seasonally dry tropical forest (SDTF) characterization, aiming to point out weaknesses and strengths of this methodology in such unique environments. The 26 SDTF in Brazil, called Caatinga, has two distinct seasons: dry and rainy. The vegetation cover among seasons changes drastically, as well as the animal activity. The reproductive period of birds, anurans, mammals and insects in the area is highly dependent on the water supply (Santos et al., 2011). The objective of this study was to describe the Brazilian SDTF soundscape dynamics and track diel patterns and seasonal changes. Specifically, we aimed to answer: (1) Which indices can better describe SDTF acoustic activity, considering multiple taxonomic groups? (2) Are there diel patterns? (3) Which animal groups contribute to changes in the observed diel patterns? (4) Do acoustic patterns change between rainy and dry seasons? We hypothesize that a single index will not be able to reflect all acoustic activity, given the diversity of signals emitted by animals. We expect that in such tropical environments, birds will dominate the diurnal period and insects will fill in the nightscape. Anurans are not expected to be present once the recording sites were not close to water bodies (information about the sampling sites are in Appendices, Table A.1). We also expected less contribution of avian fauna and a shorter dawn and dusk choruses during the dry season, as a response to climate. Regarding insects, we do not expect changes in the daily activity period as seasons change, but we do expect that the number of individuals, and even species, will reflect more intense use of the acoustic space during the wet season, when most species are exploiting the abundance of water as a resource for adults and early stages of insect’s life cycles. 2. Methodology 2.1. Study area Seasonally Dry Tropical Forests represents 20% of world tropical forests (Hansen et al., 2010), are distributed within Americas, Africa, Eurasia and Australasia (Miles et al., 2006) and 27 share some similarities such as a significant numbers of endemic species which are susceptible to desertification processes, and are often absent from priority conservation actions (Miles et al., 2006; Santos et al., 2011). Within Brazil, the SDTF is distributed over nine states in an area of 836,001 km² (MapBiomas, 2019). Annual rainfall ranges from 200-800 mm and the rainy season lasts for 3 to 5 months (Olmos et al., 2005). The study area is located within Serra do Feiticeiro (Figure 1) which is the most pristine Caatinga area in Rio Grande do Norte State (Cordero-Schmidt et al., 2018; Marinho et al., 2018; Vargas-Mena et al., 2018). Besides the conservation importance, there is no legal preserve in or around the area and a massive windfarm has been planned to be constructed in the next few years. Therefore, this paper also aims to contribute rare prior baseline data to monitor how the windfarm construction will influence the soundscape in the future (Dahl et al., 2012). 2.2. Data acquisition In each of the five sampling points, we installed one autonomous recorder (Song Meter SM3®, Wildlife Acoustics, Inc., Concord, Massachusetts), with an omnidirectional waterproof microphone, attached to a tree at 1.5 m high. The sensors were programmed to record one minute every 15 minutes at a sample rate of 44.1 kHz and 16 bits, in WAVE format. Data were collected in 2017 over 26 consecutive days each season simultaneously in each of the five sites (Figure 1). Additionally, two SM2 with the same specifications were used during the 2018 rainy season, also recording for 26 days, in two other sites simultaneously (Figure 1). Due to equipment malfunction, data from one of the recorders from the 2017 rainy season was removed from analyses. The recorders were distributed so that our sample was representative of the landscape in the region. Sites varied in vegetation structure, distance 28 from roads, terrain inclination. Distance between sampling points varied from 200 to 2,000 meters (Figure 1). Additional information about sampling sites are presented in Appendices (Table A.1 and Figure A.1). Figure 1 - Location of sites within an area of Caatinga sampled using song meters distributed in the municipality of Lajes, Rio Grande do Norte State, Brazil and a close-up photograph of one SongMeter 3 installed in the area. 2.3. Data processing To better understand the biological content of the data, we manually counted sonotypes in a subsample of the recordings (1-minute every 30 minutes on files from 3 consecutive days, for each season, N= 290). We define “sonotype” as a note or series of notes with a distinct pattern that may represent a species vocalization (Aide et al., 2017; Ferreira et al., 2018). The use of this approach has some caveats, although for insects and amphibians 29 one sonotype equals one species, in birds we have a different scenario. The same species can have a wide vocal repertoire, including mimetic signals from other species, which may lead us to biases in the number of species resulting from this approach. Nonetheless, it is valid since the limitations to identify sounds to the species-level is well understood (Aide et al., 2017) and our aim was not identifying species. Insects are currently the most problematic taxa in terms of species identification. The challenge and weakness of this methodology for soniferous insect species is that their vocal parameters have a strong association with body size and environmental conditions such as humidity and temperature (Robinson & Hall, 2002). Thus, a single insect species comprised by individuals with size variation could be accounted for two or more times. Sonotypes were detected using manual inspection of files for their spectral features in Raven Pro 1.5® (Cornell Lab of Ornithology, Ithaca, NY). Each sonotype was placed in one of four categories: birds, insects, anurans, or mammals. We also registered the presence of geophony (e.g. wind, rain) and antropophony (e.g. vehicles, gunshots). Once the recorders are exposed to adverse conditions in the field, it is common that some files are damaged (Sankupellay et al., 2015). Although it is also a common practice to manually remove files with too much wind, rain or anthropogenic noise. Instead here we use all files since we consider that those sound sources are also part of the soundscape and should be accounted for. 2.4. Acoustic Indices The acoustic indices calculations and visualization were generated using Analysis Program (Towsey et al., 2016). In total, we use 14 Summary and six Spectral Indices (see section “Data visualization” below) (Table 1). The summary indices represent a single statistic 30 calculated across a broad frequency range to measure different aspects of the acoustic energy within that time-period, in our case, 30 seconds. Meanwhile, the spectral indices are vectors calculated within predefined frequency bands and time periods, but also summarise several aspects of the acoustic energy distribution (Towsey, 2017). Table 1. Indices calculated using the Analysis Program. For details about calculations, see references. Note that ACI, ENT, EVN, BGN, PMN, R3D were calculated at both spectral and summary versions. The spectral calculations were used to build False Colour Spectrograms. ACI Acoustic Originally developed to reflect bird activity, excluding Pieretti, Complexity constant and low frequency sounds (like human Farina, & Index generated noise). Very sensible to other sound sources, Morri, 2011; like rain. In this study, based on bird species observed, Towsey, 2017 we chose to use the frequency band of 0.5 to 11 kHz. ENT Temporal Measurement of acoustic energy concentrated in each Towsey, 2017 Entropy frequency bin (Spectral Temporal Entropy), across the wave envelope. EVN Events per Average number of times the decibel envelope crosses a Towsey, 2017 second BGN +3 dB threshold, per second. BGN Background Noise profile calculated from the decibel waveform. Towsey, 2017 noise Note that this index is used to calculate others (like EVN). SNR Signal to Difference between the BGN value and the maximum Towsey, 2017 noise Ratio value in the decibel envelope. ACT Activity Fraction of values in the 30 seconds decibel envelope Towsey, 2017 that exceed 3 decibels above the BGN value. Constant sounds with high signal to noise ratio (like cicada chorus) are reflected in high ACT values. HFC High Fraction of the noise-reduced decibel spectrogram in the Towsey, 2017 Frequency high frequency band (11-22.050 kHz) that exceed 3 dB Cover above BGN. MFC Mid Fraction of the noise-reduced decibel spectrogram in the Towsey, 2017 Frequency mid frequency band (0.5-11 kHz) that exceed 3 dB above Cover BGN. LFC Low Fraction of the noise-reduced decibel spectrogram in the Towsey, 2017 Frequency low frequency band (below 0.5 kHz) that exceed 3 dB Cover above BGN. EAS Entropy of Concentration of mean energy in the mid-band of the Towsey, 2017 Average mean energy spectrum. Spectrum 31 EPS Entropy of Concentration of spectral maxima values in the mid Towsey, 2017 the Spectral frequency band (0.5 – 11 kHz). Peaks ECV Entropy of Similar calculation to EAS, but in ECV the mid-band Towsey, 2017 Coefficient spectrum is derived from the variance divided by the of Variation mean of the energy values in each frequency bin. CLC Cluster Measure of the degree of internal acoustic structure, or Towsey, 2017 Count spectral diversity, within the mid-band. Therefore, it is an index that can reflect bird diversity. SPD Spectral A measure of the number of cells in the mid-frequency Towsey, 2017 Peak Density band that are identified as being local maxima. Not normalized to be independent of frame size and frame overlap. R3D Three Ridge Combination of maximum values of three ridge indices Towsey, 2017 Indices (horizontal, upward slope, downward slope) which attempts to detect harmonic structures in the mid-band. 2.5. Data visualization Sound visualization is an important tool considering large datasets and it is based on the assumption that the human pattern perception is better at integrating long visual representations than long aural stimuli (Sankupellay et al., 2015). In this sense, the use of False Colour Spectrograms (FCS) capitalizes on the human capability to discriminate red, green, and blue and is generated based on a combination of acoustic indices. Three different and non-redundant indices are assigned to one of three colours (red-green-blue, RGB). The pixel size depends on the sampling rate (vertical axis) and temporal window (horizontal axis). Six acoustic indices were used in two different visualizations: ACI-ENT-EVN and BGN-PMN- R3D. 2.6. Clustering the dataset The k-means is one of the most popular and relatively simple clustering algorithms and requires that a fixed number of clusters is determined prior to the analysis. The algorithm used here was a variation of the k-means, known as k-means++, that results in more accurate 32 centroids and minimizes the distance between points within each cluster (Campbell et al., 2006). To determine the ideal number of clusters we used a smaller dataset with high biophonic activity. Days with rain, strong winds and recorder malfunction were discarded by inspecting the False Colour Spectrograms. This smaller dataset consisted in eight days, two for point 02 and two on point 03, in the rainy and in the dry season, respectively. The indices were normalised between 2 and 98 percentile bounds and scaled between 0 and 1. We performed a correlation matrix to remove highly correlated summary indices (Pearson’s correlation > 0.75) (Gage et al., 2017). In total, we used 11 of the 14 summary indices (Appendices, Table A.2). We tested two indices that are metrics for evaluating clustering algorithms: Silhouette Index (Rousseeuw, 1977; R Package ‘cluster’) and Dunn Index (Dunn, 1974; R Package ‘clValid’). As observed by Phillips et al. (2018), although these are the most used indices for this purpose, they do not always work. The Silhouette Index achieved best values for our sample with 5 clusters, which is a low number to represent complex acoustic communities (Phillips et al., 2018), and the Dunn Index showed incongruent results, with best values in 15 and 50 clusters. Considering the caveats of such validation metrics, we decided to manually browse 376 files and labelled sonotypes as: Orthopteran, Cicada, Birds, Wind, Quiet, Clipping, Frogs, Donkey, Anthropophony. Within Orthopteran, Cicada and Birds we also created a “Low activity” subcategory which was used when the acoustic signals were faint (low signal to noise ratio) or appeared only sporadically in the file. When k=50, the number of clusters assigned to each manually labelled class reached a plateau, except for Orthopteran, that kept increasing. An explanation for this increase may be the great number of files containing Orthopteran vocalization, and not necessarily an effect of the number of clusters. Therefore 33 we used the optimal number of clusters as 50 which was coherent with one of the Dunn Index best values. The complete dataset comprises more than 24,000 one-minute files from all seven different sampling points. These files were further divided into 30-seconds segments to achieve a better resolution and avoid losing data due to clipping. Files were removed from further analyses by excluding those with Clipping Index values >500. The Clipping Index is a measure of the proportion of audio segment that presented clipping, which compromises the information, indices calculation and, thus, would affect the clustering. We normalized the rest of the data and excluded the correlated variables (as explained above), then we performed a k-means++ clustering, using k=50, maximum number of iterations =500 (Campbell et al., 2006; R Package ‘LICORS’). We validated the content of clusters by manually verifying 10 files within each cluster, in a total of 500 files. Each file was labelled with one to four classes (birds, orthopterans, cicadas, wind, rain, quiet) which were further ranked by importance (i.e., the amount of occupation of the acoustic space). Differences in cluster categories frequencies among seasons were verified using a Chi-Squared test. 2.7. Clusters visualization We chose the Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) for data visualization (Figure 5). UMAP is a new technique for data reduction, and presents some advantages compared to other popular data reduction algorithms, such as PCA and SammonMap (McInnes et al., 2018). The UMAP analyses were generated in R (package ‘umap’) and the graphs were created using package ‘ggplot2’. 34 2.8. Relationship between clusters and acoustic indices To understand the weight of acoustic indices on each cluster we use radar plots analysis (R package ‘fmsb’), for each one of the 25 clusters associated with a single label (Figure 6). Besides using the normalized values, we also reduced the noise by using a medoid value for each variable (Phillips et al., 2018). This way, we could easily visualize which variables have the most influence on each cluster. 2.9. Diel and seasonal patterns To understand how the diel and seasonal patterns could be identified using the clusters, we chose 5 days per season in each sample point and looked for the relationship between the time of the day (24hs) and the number of files associated to each sound label. We also used the most important acoustic indices from the radar plots as a measure of the contribution of each one of our main biophonic labels. 3. Results 3.1. Visualization using False Colour Spectrograms (FCS) We compared the spectrograms (examples can be found on Appendices, Figure A.2) with the FCS to verify which sound sources were contributing more to the patterns observed (Figure 2). In a 24h resolution, it is possible to visualize the nocturnal insects at the grey scale spectrogram, but not the diurnal biophonic activity, which are easily observed in the FCS. This 35 is especially evident for short duration signals, the case for most bird calls, and an example can be seen in the dawn chorus (indicated with red arrows in Figure 2). Figure 2 – 24-hour visualization, using one-minute samples every 15 minutes. A total of 96 files were used to compose these images, no gaps between them. On top, two False Colour Spectrograms, using three different indices to compose each of the two Red-Green-Blue visualizations. The indices and 36 colours used in each FCS are indicated beside each image. In the bottom, a greyscale spectrogram of the same period, FFT=1024, 50% overlap. Red arrows indicate the dawn chorus. Although the bird activity is visible during twilight and daylight (from 04h30 to 17h30 approximately), the characteristics of the sounds produced by this group are not easily visualized and individualized with the FCS technique in our recording scheme (Figure 3). On the other hand, insects presented clear patterns of occupation of acoustic space and were better visualized in the FCS, since their vocalizations occupy narrower frequency bands with long durations. In a first look, it is even possible to estimate the number of sonotypes by counting different bands. The activity at night was dominated by Orthopterans, but it is also possible to identify Cicada’s calls as horizontal bands during the day between 5 and 7 kHz in the rainy season (Figure 3). The background noise representing the low intensity wind can be observed as the reddish colours (Figure 3, bottom FCS). Strong winds directly hitting the microphone as seen with a distinct pattern of green and were a limiting factor to the use of FCS during the dry season (vertical stripes on the FCS, Figure 4). On the other hand, this visual analysis makes it possible to detect a decrease in animals´ acoustic activity in the dry season compared to the rainy season. 37 Figure 3 - Examples of 24-hour soundscape visualizations of a seasonally dry tropical forest at the Municipality of Lajes, Rio Grande do Norte State, Brazil using False Colour Spectrograms. Three days in the rainy season (top) and three days in the dry season (bottom). Top images (for both seasons) were made using the indices ACI-ENT-EVN. Bottom images (for both seasons) were made using the indices BGN-PMN-R3D. 38 Figure 4 – Comparison between False Colour Spectrograms and graphs with the number of sonotypes manually identified for each animal category. Note that during the dry season the wind dominates the FCS visualizations and impairs inferences on acoustic activity diel fluctuations. FCS were built with the combination of BGN-PMN-R3D. 3.2. Data Clustering The number of clusters labelled as each of the categories (Birds, Cicadas, Orthopterans, Quiet and Wind) can be seen on Table 2. The first part of the table lists only the clusters assigned to a single label. During the dry season, more than 70% of the clusters fit into this single label-category, but during the rainy season, the percentage drops to 50%. In both seasons, the clustering was more successful in grouping files with Wind, Orthopterans and with very low acoustic activity (Here referred as Quiet). The frequency of all categories, except Birds+Quiet, were significantly different between seasons. 39 Table 2 – Number of clusters and files assigned to each label. The clusters are divided in those assigned to only one label, to two labels and, three or more/inconsistent labels. Percentages are shown within parenthesis. Significant differences between seasons are indicated as *(p<0.001). Number Files assigned (%) Label of Dry Rainy clusters Birds * 1 0 637 (2.56) Cicadas * 1 204 (0.82) 5 (0.03) Orthopteran * 10 1793 (12.72) 6155 (24.8) Quiet * 5 3691 (26.18) 2301 (9.27) Wind * 8 4545 (32.24) 3358 (13.52) Total 25 10233 (71.96) 12456 (50.6) Birds+Orthopteran * 4 246 (1.74) 2246 (9.04) Birds+Wind * 3 1749 (12.4) 1434 (5.77) Birds+Quiet 1 317 (2.24) 455 (1.83) Orthopteran +Cicadas * 2 225 (1.6) 770 (3.1) Orthopteran + Quiet * 1 214 (1.51) 513 (2.06) Orthopteran + Wind * 3 97 (0.68) 1970 (7.93) Birds + Rain * 1 66 (0.46) 206 (0.84) Wind + Rain * 1 0 831 (3.34) Total 16 5828 (20.63) 8425 (34.22) Birds+Orthopteran+Wind * 4 292 (2.07) 2043 (8.29) Inconsistent * 5 856 (6.07) 1694 (6.82) Total 9 1148 (8.1) 3737 (13.7) The distribution of the clusters is shown in Figure 5. We chose to show only the files assigned to a single label, once the visualization of clusters assigned to multiple labels jeopardizes interpretation of the results due to the great number of different colours used. The clusters distribution grouped together the categories Wind and Quiet, as expected. Also, we labelled as Wind, files with different intensities of wind. Thus, files with presence of low wind could have been confounded with those labelled as Quiet. Clusters labelled as Orthopterans varied in activity level, which resulted in the presence of this label in different groups of clusters. The group located in the upper right corresponded to the low activity level, mainly in the dry season, but also in late night hours in the rainy 40 season (an UMAP with files coloured as dry and rainy season is shown in Figure A.3). Files labelled as Birds were segregated from most of the other files but presented some similarities with part of the files labelled as Orthopteran during the rainy season. Figure 5 – Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) Clusters visualization. The colours represent single-label files. 3.3. Relationship between clusters and acoustic indices The weight of indices in each cluster was explored using radar plots (Figure 6). Orthopteran clusters presented two different patterns, one linked to EPS (Entropy of Peaks Spectrum) and EAS (Entropy of Average Spectrum) and other highly influenced by HFC (High Frequency Cover). This last one reflects acoustic patterns of the Tettigoniidae family, which produce high frequency sounds (Montealegre-Z and Mason, 2005). The clusters labelled as Quiet, Wind, and Cicada presented high values of Background Noise (BGN). EPS, although important to describe Orthopteran activity, showed the highest values in the clusters labelled as Birds. 41 Figure 6 - Radar plots showing the indices weights in clusters associated with a single label. Here, the medoid values were used to achieve a better visualization. BGN (Background Noise), CLS (Cluster Count), ECV (Entropy of Coefficient of Variation), EPS (Entropy of Peaks Spectrum), EAS (Entropy of Average Spectrum), ENT (Temporal Entropy), ACI (Acoustic Complexity), LFC (Low Frequency Cover), MFC (Medium Frequency Cover), HFC (High Frequency Cover), EVN (Events per Second). More details about the indices are presented in Table 1. 3.4. Diel and Seasonal Patterns Finally, we used two methods to investigate the diel patterns of our main biophonic sources contributing to the soundscapes (Figure 7) and how these patterns vary between seasons. In graphs A to D we used the number of files labelled to each animal taxa according to the clustering presented previously. Figure 7 shows similar general patterns of biophonic activity as those generated by manual identification (Figure 4). Figure 7 also shows that acoustic activity from insects and birds are almost mutually exclusive, showing little temporal overlap, markedly in the dry season. Figure 8 shows the temporal dynamics of three indices 42 that are most representative in the clusters labelled as Birds (EPS), mid-frequency band Orthopteran (EAS) and high-frequency band Orthopteran (HFC). It is interesting to observe that Orthopterans belonging to the Tettigoniidae family (HFC) had a distinct temporal pattern compared to other Orthopterans, being active only at dusk in the rainy season. EPS has a strong association with both Birds and Orthopteran clusters. Therefore, the diel pattern of EPS is different between seasons, following the acoustic activity of Birds in the dry season, but also influenced by Orthopteran activity in the rainy season. Figure 7 - Diel patterns (shaded area represents nigh time) of acoustic activity presence of the most representative groups (Orthopteran and Birds) during Dry and Rainy seasons. A 43 and B refer to files labelled as Birds, and C and D to the ones labelled as Orthopteran throughout 24 hours, using the average number of files of a subsample of five consecutive days from each of the four recorders. Figure 8 – Fluctuations in the mean value of three different indices along the 24 hours, each one associated to one biophonic group: EPS (Entropy of Spectral Peaks, associated to Birds), HFC (High frequency Cover, associated to Tettigoniidae orthopterans), and EAS (Entropy of Average Spectrum, associated to other Orthopteran families). We used the average of subsample of five days from four recorders. 4. Discussion 4.1. Acoustic Indices Here, the use of multiple acoustic indices achieves good results in tracking the daily and seasonal activity patterns in the soundscape. Previous works in similar biomes using single acoustic indices were able to use them to find variation in the soundscapes and acoustic activity between seasons, forest cover (Rankin & Axel, 2018), impact from open mining (Duarte et al., 2015), and distance from roads (focusing on birds, Machado et al., 2017). With a focus on multiple taxa, four acoustic indices (Acoustic Diversity, Evenness, Entropy, and Normalized Difference Soundscape) were correlated with insects’ richness, but the relationship with bird acoustic activity wasn’t clear (Ferreira et al., 2018). Our approach uses 44 a combination of indices to produce more reliable characterization of acoustic signals, allowing us to differentiate sound sources on the environment more precisely. The use of multiple indices has been suggested to increase the reliability of these metrics in environmental monitoring (Eldridge et al., 2018; Towsey et al., 2014). 4.2. Visualization through FCS The use of False Colour Spectrograms has some advantages based on the human capabilities of rapidly processing visual information. FCS provide a quick and reliable look at the data in its entirety and show daily acoustic patterns, also enabling the identification of the main factors responsible for the Caatinga acoustic fingerprint and the variation between its two defined seasons: dry and rainy. Visible acoustic activity patterns were different between seasons, especially when comparing the two main sources of biophony - birds and insects. These animals showed higher acoustic activity in the rainy season, as expected, considering the reproductive period of these groups is linked to the food and water availability during the wet season (Cavalcanti et al., 2016; Wolda, 1988). The observed temporal segregation in acoustic activity between these two faunal groups suggest temporal partition of the soundscape but we can also speculate about acoustic crypsis of insects during birds´ activity period, or acoustic avoidance (as coined by Curio, 1976). Acoustic avoidance may denounce a possible ecological prey- predator relationship and/or evolutionary processes of predation avoidance at play. Bird guilds pure-insectivorous, insectivorous-frugivorous, and insectivorous-granivorous account for 78% of all bird guilds in the region (Cavalcanti et al., 2016), which make this speculation plausible. 45 The wind was the main factor affecting data integrity and signal detectability, which was also expected. Wind is constant at the Serra do Feiticeiro, which also attracts Windfarms Companies to the region, where one of the biggest windmill complexes is expected to be built in the next few years. The wind (and resulting overload and clipping in the recordings) was more pronounced in the Dry season when vegetation structure - leafless trees and no herbs - allows wind to directly hit the microphones. Strong winds might also act as a masking agent of acoustic communication and can influence negatively in the decisions to call during windy periods, which would in turn affect the dynamics of the soundscape. Removing all those files was an option, but we believe that it would not realistically depict the sampled soundscapes. Additionally, we wanted to take on the challenge of investigating soundscape sampling in areas with open vegetation, as dry forests and savannas. The potential to identify groups or even species using FCS was explored by Towsey et al. (2018) and Indraswari et al. (2018). Acoustic space partitioning among insect species is clearly observed in FCS of SDTF at a much greater temporal scale that usually done (Schmidt & Balakrishnan, 2014). While vertical stratification of insects assemblages is often observed in areas with high canopy (Jain & Balakrishnan, 2011), our area is dominated by shrubs and small trees. Therefore, we suggest that the main strategy for acoustic masking avoidance among insects is frequency partitioning. In this sense, the FCS can allow studies of acoustic niche partitioning at a broader scale. To the best of our knowledge, this was the first study that employed FCS in non- continuous recordings and produced very satisfactory visualizations. The use of non- continuous recordings schemes has logistic advantages (i.e., reduction of costs related to data acquisition and storage, longer duration of passive acoustic monitoring efforts, Sousa-Lima et 46 al., 2013). Disadvantages include the reduction in the sampling resolution and loss of opportunities to detect target species that are rare or less vocal. However, to make more general inferences and comparisons across soundscapes and their main sound sources, this methodology deemed successful. 4.3. Clustering evaluation We evaluated how effective was clustering of the summary indices by listening to a sample of each cluster, examining the False Colour Spectrograms, UMAP visualization and investigation of daily patterns. Additionally, we determined how the main environmental sound sources affected the indices, inferring which index best predicts the occupation of acoustic space by each sound source. Entropy of Spectral Peaks (EPS) is a measurement of concentration of spectral maxima in the mid-band (Towsey, 2017), in our case from 0.5 to 11 kHz, a frequency band occupied by Birds, but with influence from insect stridulation as well. The index presented higher values in clusters classified as containing acoustic activity from Birds but was also present in Orthopteran clusters. The 24h variation observed in Figure 8 is a reflex of this finding. In the dry season, when Orthopteran activity is low, the EPS values increase during the day, in a pattern that resembles the Birds Activity pattern. On the Rainy season, when Orthopteran are highly active, they appear as the main driver to both EPS and EAS daily variation. Entropy of Average Spectrum (EAS) was highly correlated to Orthopteran acoustic activity. The index is a measure of the concentration of mean energy within the mid-band of the mean-energy spectrum. These insects were predominant within the mid-band, particularly at night, sometimes filling almost the entire mid-band frequency. Nonetheless, other clusters with high Orthopteran acoustic activity showed a different pattern, with high 47 values of High Frequency Cover (HFC). This index is a measure of the energy concentration on high frequency bands (here, from 11-22 kHz), assumed to be indicating acoustic activity from Tettigoniidae. The Background Noise (BGN) was highly correlated with two of our cluster categories, Quiet and Wind. This index is calculated from the energy distribution in the waveform envelope and was expected to be high on files with low (or no) biophony. Within the same taxa some similarities in the sound characteristics are expected, reflex of physiological constraints, such as sound production mechanisms. For example, all around the world, most orthopterans will produce sound by stridulating (Montealegre-Z & Mason, 2005; Riede, 1998), resulting in similar sound patterns. On the other hand, acoustic communities are highly dependent on their species composition and other environmental factors that may lead to a change in the use of acoustic space. Sound sources identified in our work didn’t affect the summary indices the same way that the ones identified by Phillips et al. (2018) in two Australian forests. Compared to Phillips et al. (2018), insect clusters also presented high values of EAS, but the authors did not find the HFC related to these animals. Besides the differences in insect assembly, this can also be a result of the sampling frequency used and the values applied to be considered as Low, Medium and High frequency in calculations. The HFC was more related to cicada activity. The bird clusters also presented differences, being more related to MFC, LFC and even ACI, while in our work, was more related to EPS. 4.4. Insects The disproportional occupation of acoustic space by insects in our data brings to our attention the importance of considering this group in ecoacoustics studies. Especially in semi- arid environments, insects represent an important food source for other animals and, 48 although there is a consensus that they are good bioindicators (Brown, 1991), there is limited evidence to support this assumption. In humid tropical environments, there is almost no variation between the insect abundance through the seasons, but a reduction can be observed in tropical environments with a severe dry season, as the Brazilian SDTF. This reduction can be caused by several factors as stress related to food decline, which can lead to adaptations that include dormancy, diapause or migration (Pinheiro et al., 2002). The use of acoustic monitoring has the potential to become a key tool in the study of insects communities (Lehmann et al., 2014), even accelerating the discovery of new species. Besides that, there is a lack of consistent and reliable acoustic libraries for the Insect group. For example, there are at least ten thousand Orthoptera species that produce sound as a form of communication, but only 10% of those are included in digital acoustic libraries (Riede, 2017). The temporal vocalization patterns in Orthoptera are variable and dependent on environmental features - especially humidity and temperature. These patterns can be continuous through the day, have peaks on dusk , peaks on dawn, only during the night, only during the daylight (Fischer et al., 1996). In tropical regions, it is common to have a nocturnal pattern, as observed in our Caatinga data. Predator avoidance is probably one of the major factors leading to this acoustic pattern. Orthopteran species living in open vegetation (as the SDTF) are easily visualized by birds, thus are more nocturnal, in comparison to species living in dense vegetation (Robinson & Hall, 2002). Changes in frequency features related to temperature changes through the night were also visible in our FCS. This phenomenon was 49 observed by Phillips et al. (2018). The temperature does not affect the frequency directly, but the wave sound speed in the medium (Endler et al., 2000). The limitation in the use of passive acoustic monitoring to study insects relies on the fact that information degradation in these animals can occur at short distances. The interval between pulses in orthopterans calls is an important information which could be lost in distances as short as 2 meters, once the echoes derived from the signals fill the spaces between them (Simmons, 1988). When we explore the greyscale spectrograms of our data, we observe that only part of the orthopteran signals presents resolution enough to make posterior comparisons with databases to identify species. 4.5. General conclusions about the use of PAM in SDTF The use of Passive Acoustic Monitoring is a relatively new technique applied to terrestrial environments and there are important open questions to be solved. The relative distance between sampled sites depends on the recording system sensitivity, but also on the signal recorded. Here we include constraints as amplitude, frequency range, position of emitter, vegetation, topography, temperature, and humidity (Farcas et al., 2016). An effective sampling relies, thus, on the coverage of different species, their vocal characteristics, and should also vary according to time and space (Gibb et al., 2018). The development of the acoustic ecology field with improved processing and analyses of time series can provide much information about the environment to managers, ecologists and conservationists (Burivalova et al., 2019; Almo Farina & Gage, 2018). We believe that such efforts are necessary in regions with low scientific knowledge and low financial investment in research and conservation, as the Brazilian Caatinga (Santos et al., 2011) and other tropical dry forests. 50 Our main challenges working with acoustic data in a dry forest was the constant presence of wind, especially in the dry season, when vegetation loses foliage. Also, the strong presence of insects associated with the relatively low abundance of birds compared to other tropical regions (again, more evident on the dry season) steers our attention to this group, that is still left aside in most ecoacoustics studies (Aide et al., 2017). Some acoustic indices are dependent on the frequency band selected for inclusion in the calculations (i.e., the frequency range is adjustable) which can bias the interpretation of the soundscape. We highlight the importance of considering the insects in both experimental design and analyses. Despite these challenges, our approach was able to track seasonal changes and patterns of animal activity, reflected in the use of acoustic space. This paper represents a contribution to a better understanding of the soundscape dynamic in a highly seasonal tropical environment. Soundscape ecology and its translation into acoustic indices to study ecological processes in tropical environments is still very incipient. Seasonally dry tropical forests (SDTF) are globally threatened and receive less economic intake in conservation initiatives when compared to other tropical environments (Dirzo et al., 1995; Santos et al., 2011), which makes even more important that fast and cheap methodologies to study biodiversity are explored. Acknowledgements We thank Marina Scarpelli, Michael Towsey, Yvonne Phillips, Miriam Pinto, Mauro Pichorim for providing comments and suggestions on this manuscript. For field work assistance we thank Eugenia Cordero-Schmidt, Juan Vargas-Mena, João Bernardino de Lima, Gabriel Sabino. For assistance with the data analysis we thank Anthony Truskinger, Philip 51 Eichinski, Michael Towsey, Lara Lopes, Heloise Ferreira. This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001, FAPESP (2013/50421-2). E. G. Oliveira was supported by Coordenação de Aperfeiçoamento de Pessoal de Nivel Superior (CAPES) with Ph.D. scholarship. 5. References Aide, T. M., Hernández-Serna, A., Campos-Cerqueira, M., Acevedo-Charry, O., & Deichmann, J. L. (2017). Species richness (of insects) drives the use of acoustic space in the tropics. Remote Sensing, 9(11). https://doi.org/10.3390/rs9111096 Blumstein, D. T., Mennill, D. J., Clemins, P., Girod, L., Yao, K., Patricelli, G., … Kirschel, A. N. G. (2011). Acoustic monitoring in terrestrial environments using microphone arrays: Applications, technological considerations and prospectus. Journal of Applied Ecology, 48(3), 758–767. https://doi.org/10.1111/j.1365-2664.2011.01993.x Bradbury, J. W., & Vehrencamp, S. L. (1998). Principles of Animal Communication (1st ed.). Sinauer Associates Inc. Brown, K. S. (1991). Conservation of Neotropical Environments: Insects as Indicators. In The Conservation of Insects and their Habitats (pp. 349–404). The Royal Entomological Society of London. https://doi.org/10.1016/b978-0-12-181370-3.50020-8 Burivalova, Z., Game, E. T., & Butler, R. A. (2019). The sound of a tropical forest. Science, 363(6422), 28–29. https://doi.org/10.1126/science.aav1902 Buxton, R. T., Agnihotri, S., Robin, V. V, Goel, A., Balakrishnan, R., State, C., Pradesh, A. (2018). Acoustic indices as rapid indicators of avian diversity in different land-use types in an Indian biodiversity hotspot. Journal of Ecoacoustics, 2(#GWPZVD), 1–17. Campbell, G., Koutinas, A., Wang, R., Sadhukhan, J., & Webb, C. (2006). k-means++: The advantages of careful seeding. Chemical Engineer, 8(781), 26–28. https://doi.org/10.1145/1283383.1283494 Cavalcanti, L. M. P., de Paiva, L. V., & França, L. F. (2016). Effects of rainfall on bird reproduction in a semi-arid Neotropical region. Zoologia, 33(6), 1–6. https://doi.org/10.1590/S1984-4689zool-20160018 Cordero-Schmidt, E. S., F. D. A. R., Vargas-Mena, J. C., Medellín, R. A., Herrera, B. R., Venticinque, E. M., Oliveira, P. P., & Barbier, E. (2018). Natural History of the Caatinga Endemic Vieira’s Flower Bat, Xeronycteris vieirai. Acta Chiropterologica, 19(2), 399–408. https://doi.org/10.3161/15081109acc2017.19.2.016 Curio, E. (1976). The ethology of predation. New York: Springer. 52 Dahl, E. L., Bevanger, K., Nygård, T., Røskaft, E., & Stokke, B. G. (2012). Reduced breeding success in white-tailed eagles at Smøla windfarm, western Norway, is caused by mortality and displacement. Biological Conservation, 145(1), 79–85. https://doi.org/10.1016/j.biocon.2011.10.012 Dirzo, R., Young, H. S., Mooney, H. A., & Ceballos, G. (Eds.). (2011). Seasonally dry tropical forests: Ecology and Conservation. Washington: Island Press. Duarte, M. H. L., Sousa-Lima, R. S., Young, R. J., Farina, A., Vasconcelos, M., Rodrigues, M., & Pieretti, N. (2015). The impact of noise from open-cast mining on Atlantic forest biophony. Biological Conservation, 191(August), 623–631. https://doi.org/10.1016/j.biocon.2015.08.006 Dunn, J. C. (1974). Well-separated clusters and optimal fuzzy partitions. Journal of Cybernetics, 4(1), 95–104. https://doi.org/10.1080/01969727408546059 Eldridge, A., Guyot, P., Moscoso, P., Johnston, A., Eyre-Walker, Y., & Peck, M. (2018). Sounding out ecoacoustic metrics: Avian species richness is predicted by acoustic indices in temperate but not tropical habitats. Ecological Indicators, 95(December), 939–952. https://doi.org/10.1016/j.ecolind.2018.06.012 Endler, J., Espmark, Y., Amundsen, T., & Rosenqvist, G. (2000). Evolutionary implications of the interaction between animal signals and the environment. Animal Signals: Signaling and Signal Design in Animal Communication, 11–46. Farcas, A., Thompson, P. M., & Merchant, N. D. (2016). Underwater noise modelling for environmental impact assessment. Environmental Impact Assessment Review, 57, 114–122. https://doi.org/10.1016/j.eiar.2015.11.012 Farina, A. (2014). Soundscape Ecology. https://doi.org/10.1007/978-94-007-7374-5 Farina, A., & Gage, S. H. (2018). Ecoacoustics. (A. Farina & S. Gage, Eds.) (1st ed.). Oxford: Wiley. https://doi.org/10.1002/9781119230724 Ferreira, L. M., Oliveira, E. G., Lopes, L. C., Brito, M. R., Baumgarten, J., Rodrigues, F. H., & Sousa-Lima, R. S. (2018). What do insects, anurans, birds, and mammals have to say about soundscape indices in a tropical savanna. Journal of Ecoacoustics, 2(March), PVH6YZ. https://doi.org/10.22261/JEA.PVH6YZ Fischer, F. P., Schubert, H., Fenn, S., & Schulz, U. (1996). Diurnal song activity of grassland orthoptera. Acta Oecologica, 17, 345–364. Gage, S. H., Wimmer, J., Tarrant, T., & Grace, P. R. (2017). Acoustic patterns at the Samford Ecological Research Facility in South East Queensland, Australia: The Peri-Urban SuperSite of the Terrestrial Ecosystem Research Network. Ecological Informatics, 38, 62–75. https://doi.org/10.1016/j.ecoinf.2017.01.002 Gibb, R., Browning, E., Glover-kapfer, P., & Jones, K. E. (2019). Emerging opportunities and challenges for passive acoustics in ecological assessment and monitoring. Methods in Ecology and Evolution, 10, 169–185. https://doi.org/10.1111/2041-210X.13101 Gibb, R., Browning, E., Glover-Kapfer, P., & Jones, K. E. (2018). Emerging opportunities and challenges for passive acoustics in ecological assessment and monitoring. Methods in 53 Ecology and Evolution, 2019(April 2018), 169–185. https://doi.org/10.1111/2041- 210X.13101 Hansen, M. C., Stehman, S. V., & Potapov, P. V. (2010). Quantification of global gross forest cover loss. Proceedings of the National Academy of Sciences, 107(19), 8650–8655. https://doi.org/10.1073/pnas.0912668107 Indraswari, K., Bower, D. S., Tucker, D., Schwarzkopf, L., Towsey, M., & Roe, P. (2018). Assessing the value of acoustic indices to distinguish species and quantify activity: A case study using frogs. Freshwater Biology, (September), fwb.13222. https://doi.org/10.1111/fwb.13222 Jain, M., & Balakrishnan, R. (2011). Does acoustic adaptation drive vertical stratification? A test in a tropical cricket assemblage. Behavioral Ecology, 23(2), 343–354. https://doi.org/10.1093/beheco/arr191 Krause, B.; Farina, A. (2016). Using ecoacoustic methods to survey the impacts of climate change on biodiversity. Biological Conservation, 195(JANUARY), 245–254. https://doi.org/10.1016/j.biocon.2016.01.013 Krause, B. (1987). The Niche Hypothesis: How animals Taught Us to Dance and Sing, 1– 6. Lehmann, G. U. C., Frommolt, K. H., Lehmann, A. W., & Riede, K. (2014). Baseline data for automated acoustic monitoring of Orthoptera in a Mediterranean landscape, the Hymettos, Greece. Journal of Insect Conservation, 18(5), 909–925. https://doi.org/10.1007/s10841-014-9700-2 Machado, R. B., Aguiar, L., & Jones, G. (2017). Do acoustic indices reflect the characteristics of bird communities in the savannas of Central Brazil? Landscape and Urban Planning, 162, 36–43. https://doi.org/10.1016/j.landurbplan.2017.01.014 MapBiomas. (2019). Coleção MapBiomas; Série anual de mapas de cobertura e uso de solo do Brasil. Retrieved September 12, 2019, from http://mapbiomas.org Marinho, P. H., Bezerra, D., Fonseca, C. R., Antongiovanni, M., & Venticinque, E. M. (2018). Mamíferos De Médio E Grande Porte Da Caatinga Do Rio Grande Do Norte, Nordeste Do Brasil. Mastozoología Neotropical, 25(2), 345–362. https://doi.org/10.31687/saremmn.18.25.2.0.15 McInnes, L., Healy, J., & Melville, J. (2018). UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. https://doi.org/10.21105/joss.00861 Merchant, N. D., Fristrup, K. M., Johnson, M. P., Tyack, P. L., Witt, M. J., Blondel, P., & Parks, S. E. (2015). Measuring acoustic habitats. Methods in Ecology and Evolution, 6(3), 257– 265. https://doi.org/10.1111/2041-210X.12330 Miles, L., Newton, A. C., DeFries, R. S., Ravilious, C., May, I., Blyth, S., Gordon, J. E. (2006). A global overview of the conservation status of tropical dry forests. Journal of Biogeography, 33(3), 491–505. https://doi.org/10.1111/j.1365-2699.2005.01424.x Montealegre-Z, F., & Mason, A. C. (2005). The mechanics of sound production in Panacanthus pallicornis (Orthoptera: Tettigoniidae: Conocephalinae): the stridulatory motor 54 patterns. Journal of Experimental Biology, 208(7), 1219–1237. https://doi.org/10.1242/jeb.01526 Olmos, F., Silva, W. A. D. G. E., & Albano, C. G. (2005). Aves em oito áreas de Caatinga no Sul do Ceará e Oeste de Pernambuco, nordeste do Brasil: composição, riqueza e similaridade. Papéis Avulsos de Zoologia (São Paulo), 45(14), 179–199. https://doi.org/10.1590/S0031-10492005001400001 Parks, S. E., Searby, A., Célérier, A., Johnson, M. P., Nowacek, D. P., & Tyack, P. L. (2011). Sound production behavior of individual North Atlantic right whales: Implications for passive acoustic monitoring. Endangered Species Research, 15(1), 63–76. https://doi.org/10.3354/esr00368 Phillips, Y. F., Towsey, M., & Roe, P. (2018). Revealing the ecological content of long- duration audio-recordings of the environment through clustering and visualisation. PLoS ONE, 13(3), 1–27. https://doi.org/10.1371/journal.pone.0193345 Pieretti, N., Duarte, M. H. L., Sousa-Lima, R. ., Rodrigues, M., Young, R. J., & Farina, A. (2015). Determining temporal sampling schemes for passive acoustic studies in different tropical ecosystems. Tropical Conservation Science, 8(1), 215–234. Pieretti, N., Farina, A., & Morri, D. (2011). A new methodology to infer the singing activity of an avian community: The Acoustic Complexity Index (ACI). Ecological Indicators, 11(3), 868–873. https://doi.org/10.1016/j.ecolind.2010.11.005 Pijanowski, B. C., Villanueva-rivera, L. J., Dumyahn, S. L., Farina, A., Krause, B. L., Napoletano, B. M.,Pieretti, N. (2011). Soundscape Ecology : The Science of Sound in the Landscape. BioScience, 61(3), 203–216. https://doi.org/10.1525/bio.2011.61.3.6 Pinheiro, F., Diniz, I., Coelho, D., & Bandeira, M. (2002). Seasonal pattern of insect abundance in the Brazilian Cerrado. Austral Ecology, (September), 132–136. https://doi.org/10.1046/j.1442-9993.2002.01165.x Rankin, L., & Axel, A. C. (2018). Biodiversity assessment in Tropical Biomes using Ecoacoustics: Linking Soundscapes to Forest Structure in a human-dominated Tropical Dry Forest in Southern Madagascar. In A. Farina & S. H. Gage (Eds.), Ecoacoustics (pp. 129–144). Wiley. Riede, K. (1998). Acoustic monitoring of Orthoptera and its potential for conservation. Journal of Insect Conservation, 2(3–4), 217–223. https://doi.org/10.1023/A:1009695813606 Riede, K. (2017). Acoustic profiling of Orthoptera for species monitoring and discovery: present state and future needs. PeerJ Preprints, (November). https://doi.org/10.7287/peerj.preprints.3397v1 Robinson, D. J., & Hall, M. J. (2002). Sound signalling in orthoptera. In P. Evans (Ed.), Advances in Insect Physiology (Vol. 29, pp. 151–278). Elsevier. https://doi.org/10.1016/S0065-2806(02)29003-7 Rousseeuw, P. J. (1987). Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics, 20, 53–65. https://doi.org/10.1177/003754977702900403 55 Sankupellay, M., Towsey, M., & Truskinger, A. (2015). Visual Fingerprints of the Acoustic Environment : The use of acoustic indices to characterise natural habitats. In IEE (pp. 111– 118). Santos, J. C., Leal, I. R., Almeida-Cortez, J. S., Fernandes, G. W., & Tabarelli, M. (2011). Caatinga: The scientific negligence experienced by a dry tropical forest. Tropical Conservation Science. https://doi.org/10.1177/194008291100400306 Schafer, R. M. (1977). The soundscape: our sonic environment and the tuning of the world. Rochester, Vt.: Destiny Books. Simmons, L. W. (1988). The calling song of the field cricket, Gryllus bimaculatus (de geer): constraints on transmission and its role in intermale competition and female choice. Animal Behaviour, 36(2), 380–394. https://doi.org/10.1016/S0003-3472(88)80009-5 Sousa-Lima, R. S., Norris, T. F., Oswald, J. N., & Fernandes, D. P. (2013). A review and inventory of fixed autonomous recorders for passive acoustic monitoring of marine mammals. Aquatic Mammals, 39(1), 23–53. https://doi.org/10.1578/AM.39.1.2013.23 Southworth, M. (1969). The Sonic Environment of Cities. Environment and Behavior, 50–70. Sueur, J., Farina, A., Gasc, A., Pieretti, N., & Pavoine, S. (2014). Acoustic indices for biodiversity assessment and landscape investigation. Acta Acustica United with Acustica, 100(4), 772–781. https://doi.org/10.3813/AAA.918757 Towsey, M. (2017). Title: The calculation of acoustic indices derived from long-duration recordings of the natural environment. Retrieved from http://eprints.qut.edu.au/61399/. Towsey, M., Wimmer, J., Williamson, I., & Roe, P. (2014). The use of acoustic indices to determine avian species richness in audio-recordings of the environment. Ecological Informatics, 21(July 2015), 110–119. https://doi.org/10.1016/j.ecoinf.2013.11.007 Towsey, M., Znidersic, E., Broken-Brow, J., Indraswari, K., Watson, D. M., Phillips, Y., Roe, P. (2018). Long-duration, false-colour spectrograms for detecting species in large audio data-sets. Journal of Ecoacoustics, 2, IUSWUI. https://doi.org/10.22261/JEA.IUSWUI Tucker, D., Gage, S. H., Williamson, I., & Fuller, S. (2014). Linking ecological condition and the soundscape in fragmented Australian forests. Landscape Ecology, 29(4), 745–758. https://doi.org/10.1007/s10980-014-0015-1 Vargas-Mena, J. C., Lima, S. M. Q., Barros, M. A. S., Alves-Pereira, K., Barbier, E., Cordero- Schmidt, E., Venticinque, E. M. (2018). The bats of Rio Grande do Norte state, northeastern Brazil. Biota Neotropica, 18(2). https://doi.org/10.1590/1676-0611-bn-2017-0417 Wolda, H. (1988). Insect Seasonality: Why? Annual Review of Ecology and Systematics, 19(1), 1–18. https://doi.org/10.1146/annurev.es.19.110188.000245 Zhang, L., Towsey, M., Zhang, J., & Roe, P. (2016). Classifying and ranking audio clips to support bird species richness surveys. Ecological Informatics, 34, 108–116. https://doi.org/10.1016/j.ecoinf.2016.05.005 56 6. Appendices Figure A.1 – Examples of areas sampled. Note the differences in top photos, taken at the same place during rainy and dry seasons. 57 Figure A.2 – Grey scale spectrograms of six distinct acoustic patterns: Gunshots (A), Orthoptera (B), Orthoptera with high frequency band stridulation (C), Birds (D), Cicada (E) and Bell (F). 58 Figure A.3 – Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) Clusters visualization. The colours represent files recorded in dry an rainy season. 59 Table A.1. General description of sampling sites used in the study. *Please note the region has an unpredictable rain regime and intermittent rivers that may flow once in every five years. To measure the nearest water body, we considered only the ones that have water most part of the year (in most cases, human-made reservoirs). Point Coordinates Habitat Elevation Nearest Nearest water (m) road (m) body (m)* 1 5°47'2.1"S Rocky ground, vegetation 224 600 2900 36°8'58.1"O composed mostly by shrubs and cactus. 2 5°47'42.1"S Terrain inclination around 268 650 2100 36° 8'3.0"O 30°, SM installed in a hill, exposed to wind guts. Vegetation include trees and shrubs. 3 5°47'49.5"S Close to a dry water course, 257 600 2300 36°8'19.9"O Terrain inclination around 30°. Vegetation composed of trees, bromeliads, and cactus. Presence of large rocks, used as shelter by bats, small mammals and lizards. 4 5°46'19.6"S Open area, used as cotton 223 320 1230 36°9'38.7"O plantation in the past, in regeneration. Vegetation dominated by shrubs, herbs and cactus. 5 5°49'56.20"S Human made reservoir. As 303 130 50 36°12'10.3"O one of the largest in the region, it contains water most part of the year. Poaching is common during the dry season in the area. 6 5°47'38.1"S Area dominated by herbs 243 30 2200 36° 8'17.6"O and sparse trees. 7 5°47'48.4"S SM positioned on top of a 325 100 2200 cave, used by bat colonies 36°14'25.6"O and other small mammal species and lizards. Vegetation is mostly composed by shrubs and small trees. 60 Table A.2. Correlation Matrix of fourteen indices for the complete dataset. Pearson’s Correlation > 0.75 are in bold. The eleven indices used after removal of the high correlated ones are presented in bold. BGN SNR ACT EVN HFC MFC LFC ACI ENT EAS EPS ECV CLS SPD BGN 1 0.420 0.254 0.205 0.342 0.313 0.313 0.120 0.370 0.071 0.523 0.137 0.386 0.398 SNR 0.420 1 0.792 0.733 0.154 0.431 0.810 0.181 0.895 0.220 0.186 0.116 0.558 0.314 ACT 0.254 0.792 1 0.828 0.172 0.449 0.866 0.140 0.618 0.177 0.200 0.121 0.456 0.330 EVN 0.205 0.733 0.828 1 0.099 0.256 0.747 0.077 0.634 0.125 0.091 0.119 0.389 0.188 HFC 0.342 0.154 0.172 0.099 1 0.520 0.196 0.319 0.107 0.040 0.136 0.029 0.423 0.878 MFC 0.313 0.431 0.449 0.256 0.520 1 0.526 0.398 0.375 0.017 0.248 0.043 0.611 0.837 LFC 0.313 0.810 0.866 0.747 0.196 0.526 1 0.175 0.708 0.153 0.239 0.100 0.496 0.388 ACI 0.120 0.181 0.140 0.077 0.319 0.398 0.175 1 0.173 0.044 0.041 0.050 0.231 0.365 ENT 0.370 0.895 0.618 0.634 0.107 0.375 0.708 0.173 1 0.202 0.131 0.113 0.498 0.256 EAS 0.071 0.220 0.177 0.125 0.040 0.017 0.153 0.044 0.202 1 0.411 0.524 0.130 0.033 EPS 0.523 0.186 0.200 0.091 0.136 0.248 0.239 0.041 0.131 0.411 1 0.095 0.247 0.261 ECV 0.137 0.116 0.121 0.119 0.029 0.043 0.100 0.050 0.113 0.524 0.095 1 0.035 0.045 CLS 0.386 0.558 0.456 0.389 0.423 0.611 0.496 0.231 0.498 0.130 0.247 0.035 1 0.639 SPD 0.398 0.314 0.330 0.188 0.878 0.837 0.388 0.365 0.256 0.033 0.261 0.045 0.639 1 61 62 Wind farm influence on the soundscape of a seasonally dry tropical forest Eliziane G. Oliveiraa,b,d, Milton Cezar-Ribeiroc, Paul Roeb, Renata S. Sousa-Limaa a Laboratório de Bioacústica (LaB), Departamento de Fisiologia e Comportamento, Biosciences Center, Universidade Federal do Rio Grande do Norte, Avenida Senador Salgado Filho, 3000, Bairro Lagoa Nova, Natal, RN. 59078-970, Brazil. b Science and Engineering Faculty, Queensland University of Technology. 2 George St, Brisbane City QLD 4000, Australia. c Spatial Ecology and Conservation lab (LEEC), Department of Ecology, State University of Sao Paulo (UNESP), Av. 24 A, 1515, Rio Claro SP, 13506-900, Brazil. d Graduate Program in Ecology, Biosciences Center, Universidade Federal do Rio Grande do Norte, Avenida Senador Salgado Filho, 3000, Bairro Lagoa Nova, Natal, RN. 59078-970, Brazil. *Corresponding author: eliziane.garcia@gmail.com Planning to submit to “Science of the Total Environment” 63 64 Abstract Passive acoustic monitoring of terrestrial environments can detect soundscape changes in temporal and spatial scales, but in tropical environments, its applicability is still limited due to the complexity of these natural systems. Most tropical systems are currently under anthropogenic influence. Here we addressed wind farm effects on a seasonally dry tropical forest in Northeastern Brazil. We aimed to understand how the sound energy distribution and acoustic diversity vary as a function of distance from wind farm turbines. We sampled seven points, four in the wind farm area, and three in a control area. Soundscape composition was described in terms of temporal and spectral patterns, using False Colour Spectrograms (FCS) and Spectral Probability Density (SPD) visualization, respectively. Two acoustic indices (Temporal Entropy and Signal to Noise Ratio) were used as response variables in Linear Mixed Models to verify if the changes in the use of acoustic space were related to the distance from the turbines. Soundscape samples were chosen to capture the activity of birds (daytime) and insects (nighttime). As we move closer to the turbines, the biophonic activity decreases, especially in the lower frequencies. Distance from turbines affected the variation in acoustic indices’ values both in the morning and at night. Our findings suggest that both birds and insects have their acoustic activity patterns modified in response to wind turbines presence, denouncing that the effect of wind farming on natural soundscapes is evident at small spatial scales. Keywords: Acoustic impact, Ecoacoustics, Wind Energy, Acoustic Indices, Acoustic monitoring 65 1. Introduction Acoustic signals are plastic and can be adjusted and modified in short periods of time according to environmental conditions (Parks et al., 2011). As a result, changes in land use can rapidly be detected acoustically by identifying loss in species richness or changes in vocalization patterns (Farina, 2014; Zwart et al., 2016). Hence, variations of temporal and spatial patterns on animal diversity can be captured non-invasively with passive acoustic monitoring (Depraetere et al., 2012; Duarte et al., 2015). Bioacoustic and soundscape analyses are powerful tools to monitor biodiversity in tropical forests (Fuller et al., 2015; Tucker et al., 2014) since such complex systems require more sampling effort using traditional methods. Soundscape analyses have been proposed as a mean to inform decision making in environmental policies (Doser et al., 2019), but few studies address its applicability and evaluate to which extent it can be used, especially in tropical environments (Burivalova et al., 2019). This acoustic approach can detect several taxonomic groups at variable distances depending on the signal and the environmental conditions (Darras et al., 2016). Eolic energy farming is a growing industry because it harvests a renewable energy source that is assumed to have low environmental impact. Environmental impact assessments indicate that the most affected taxa in terrestrial wind farm facilities are birds and bats (Anderson et al., 1999; Kunz et al., 2007). Nonetheless, most studies focus on only a single group or species and are biased towards direct mortality caused by collision with the turbines (de Lucas et al., 2012; Drake et al., 2015; Smallwood et al., 2009). 66 One additional problem associated with wind farming documented in the literature is the acoustic impact (Rabin et al., 2006; Zwart et al., 2016). Anthropogenic sound pollution is usually concentrated in lower frequency bands and can contaminate natural areas, even in the most remote places that are rarely monitored (Reed et al., 2012; Wrightson, 1999). Noise pollution can overlap with animal calls, affecting sound perception by individuals, impairing signal detection and causing behavioural alterations in receivers and senders, such as potentially costly vocal adjustments (Patricelli & Blickley, 2006; Reed et al., 2012; Slabbekoorn & den Boer-Visser, 2006). When exposure to noise is chronic, effects go beyond acoustic communication, and can induce changes in spatial distribution or reproductive traits in several species (Barber et al., 2010; Kunc & Schmidt, 2019). Although noise pollution in terrestrial habitats has been reported mainly for vertebrates (Shannon et al., 2016), there is an increasing interest in the study of its impact on insects (Duarte et al., 2019), especially in tropical habitats, where insects are most representative (Aide et al., 2017). This work was motivated by the necessity of better understanding the relationship between wind farm facilities and the acoustic community around it. Specifically, we aimed to answer the following questions: (1) How does the sound energy distribution vary along a distance gradient from turbines? (2) Do acoustic indices capture the variation observed along this gradient? (3) How can we use these measurements to infer impacts on animals living around the turbines? 67 We hypothesize that (1) sound energy (amplitude) would have high values distributed in more frequency bands in the most distant points from the noise source (turbines); (2) Signal to Noise Ratio (SNR) and Temporal Entropy Index (ENT) will capture the changes in soundscape along a distance (SNR in the amount of acoustic energy proportional to the amplitude of noise and ENT will be able to capture changes in the amplitude of short duration signals, e.g. bird calls); (3) animals with sound signals concentrated in lower frequencies in the spectrum would be more impacted, due to overlap of calls and noise in the same frequencies. To the best of our knowledge, this is the first study applying acoustic ecology to address wind farming effects and testing the application of acoustic monitoring for evaluating wind farm ecological effects in tropical environments. 2. Methodology 2.1. Study sites The wind farm facility sampled in this study is located in Jandaíra, Rio Grande do Norte State, Northeast Brazil (05° 18' 30.1" S, 36° 02' 31.4" W). The majority of this area is characterized by abandoned pastures, in varying stages of natural regeneration. Recorders were installed in a fragment where vegetation structure and plant species composition still resemble the original vegetation in the area. The area used as control is a preserved area located in the municipality of Lajes, Rio Grande do Norte State (05° 47' 42.1" S, 36° 08' 03.0" W). The vegetation structure and topography of the area are different from the wind farm, but previous inspection of the recordings revealed a similarity in the use of acoustic space and general acoustic activity patterns (see False Colour Spectrograms section for details). 68 2.2. Recordings Data collection started at the end of the rainy season in June 2017. We used five SM3 (Song Meter SM3®, Wildlife Acoustics, Inc., Concord, Massachusetts), recording for 26 days at a sampling rate of 48 kHz. One of the recorders failed and the data was not used for further analyses. We placed the SM3 recorders along a 1 km line perpendicular from the turbines, spaced 200 meters between them to ensure the independence of sampling points (Pieretti et al., 2015) (Figure 1). As a control, two recorders were placed in a more preserved area, with six km of distance between them (Figure 1). 69 Figure 1 – Location of recorders in the two study areas. On top, the wind farm area, within the municipality of Jandaíra, Rio Grande do Norte State (RN, Brazil), and on the bottom, the control area, within the municipality of Lajes, RN, Brazil. In site photos on the left are from wind farm area and show a SongMeter 3 recorder and the wind turbines. 2.3. Characterization of species and sonotypes The number of bird and insect sonotypes was measured in a subsample of 300 minutes, randomly extracted from morning (05h00 to 07h00) and evening periods (18h00 to 20h00) at the most distant point in the wind farm area. A sonotype can be described as a distinct note 70 or series of notes that represents one type of vocalization of a species (Aide et al., 2017; Ferreira et al., 2018). We chose to use sonotypes because it is difficult to identify insects to the level of species. Two ornithologists identified the bird species. From this sound database, we extracted the values of dominant frequency of all acoustic signals. A minimum of 10 samples for each species/sonotypes were used. 2.4. Sound level distribution We generated sound level distributions over the frequency spectrum using Spectral Probability Density (SPD) analyses (Merchant et al., 2013). Though there are several averaging methods for ambient noise spectra, the most common and the one we used here, is the RMS level, where the mean is computed before it is converted to dB. We used the PAMGuide function for MatLab R2018, following Merchant et al. (2015). The parameters were set to default, except Linear scale, Gain of 24 dB, zero-to-peak voltage of the analogue-to-digital converter (vADC) of 1.414. 2.5. Acoustic Indices To inspect the differences along the distance gradient from the turbines, we used two subsets of data. One, focusing on bird species, with data from 05h00 to 08h00. Other, focusing on insects, with data from 18h00 to 20h00. In both of them, minutes were sampled in a scheme of 1 minute every 15 minutes (Pieretti et al., 2015). Here, we used 15 days of recordings, the same days were sampled in both control and wind farm areas. We use the open software QUT Ecoacoustics Analysis Program (Towsey et al., 2016) to calculate the acoustic indices. Six spectral indices were used to produce the False Colour Spectrograms (ACI, ENT, EVN, BGN, PMN, R3D see Table 1). Based on inspection of the values 71 distribution of all acoustic indices, we decided to use Temporal Entropy (ENT) and Signal to Noise Ratio (SNR) to verify the differences between recording points (see Statistical Analysis for details). The ENT was used in both spectral and summary calculations. The difference between them is that, for the spectral indices, values are assigned to each frequency band. In the summary calculations, indices’ values were calculated to produce a single value for each 1-minute sample. Table 1. Description of acoustic indices used. For details about calculations, see references. The faunal groups (Birds or Insects) to which the indices are more correlated are indicated between parentheses. BGN Background Noise (Towsey, 2017): Noise profile calculated from the decibel waveform. ENT Temporal Entropy (Towsey, 2017): Measurement of acoustic energy concentrated in each frequency bin (Spectral Temporal Entropy), across the wave envelope. It is a good measure of signals that conentrate energy in short periods, such as birds. EVN Events per Second (Towsey, 2017): Average number of times the decibel envelope crosses a BGN +3 dB threshold, per second. ACI Acoustic Complexity Index (Pieretti et al., 2011): Originally developed to reflect bird activity, excluding constant and low frequency sounds (like human generated noise). Very sensible to other sound sources, like rain. PMN Power minus Noise (Towsey, 2017): The maximum decibel value in each frequency bin of the noise- reduced decibel spectrogram. R3D Three Ridge Indices (Towsey, 2017): Combination of three other Ridge Indices (Horizontal, Vertical, Downward Slope). The ridge indices attempt to detect the harmonic structure presented in calls. SNR Signal to Noise ratio (Towsey, 2017): Difference between the BGN value and the maximum value in the decibel envelope. We used the acoustic indices to build False Colour Spectrograms (FCS) (Towsey et al., 2014), in order to visualize the acoustic space occupation and daily patterns of animal groups. The FCS images are the result of a combination of three indices, represented by three colours (red-green-blue). Each of these three indices chosen detects part of the information within 72 the recording. We used the following indices to generate two different FCS: ACI-ENT-EVN and BGN-PMN-R3D. 2.6. Statistical Analyses We used R (R Core Team, 2018) and lm4 (Bates, Mächler, Bolker, & Walker, 2015) to perform a linear mixed effects analysis of the relationship between acoustic index and distance from turbines. Distance from turbines was entered as fixed effect. As random effects, we included area (windfarm and control) and time (day per sampling point). Visual inspection of residual plots did not reveal deviations from normality or homoscedasticity. P-values were obtained by likelihood ratio tests of the full model with distance against the null model. 3. Results 3.1. Signals’ characterization We obtained a total of 630 hours of recordings, including control and wind farm areas, and the sampling of continuous one-hour recordings during dawn (5h30 to 06h30) and dusk (17h30 to 18h30) periods. Continuous recordings were used to build False Colour Spectrograms and to identify bird species and insects sonotypes (in subsamples – 300 minutes total). In the wind farm area, we identified a total of 29 bird species (presented in Supplementary Material, Table S1). Bird vocalizations presented dominant frequency mean value of 3246.8 Hz (varying from 515 to 7448.9 Hz) and insect sounds presented a mean value of 7938 Hz (varying from 3188 to 18797 Hz) (Figure 2). 73 Figure 2 – Boxplot showing distribution of dominant frequencies values for birds and insects, detected in recordings at dawn and dusk hours close to a wind farm facility, municipality of Jandaíra, Rio Grande do Norte state, Brazil. 3.2. Sound level distribution To visualize the energy distribution along the frequency bands, we generated a Spectral Probability Density (SPD) plot (Figure 3). The SPD plots show the empirical probability density of sound levels in each frequency band. The general pattern of the SPD for the wind farm area presents two peaks around 5 kHz e 10 kHz (Figure 3). At the closest point to the turbines, activity around 5 kHz is less evident than in the other sample points. This is the frequency band most used by birds but is also occupied by insects (Figure 2). The frequencies around 10 kHz, on the other hand, are used almost exclusively by insects, and mainly of the signals come from one Orthoptera group – the katidids (Figure S1). Around 5 kHz we can visualize another difference among sample sites: while in the point at 100 meters there is only one peak in the RMS values (pink line, examples 74 are pointed with red arrows), the others shown two or even three peaks, suggesting a higher diversity in biophonic signals (probably due to diverse insect activity). The control area, represented in Figure 3, shows a different pattern, with no prominent peak between 10 and 14 kHz. To compare, we inspected the spectrograms (Figure S1) and detected the constant presence of one Orthopteran sonotype in the wind farm areas at night occupying these higher frequency bands. It is also interesting to note the differences in RMS peaks between 2 and 7 kHz between wind farm and control areas. The control area presented several peaks, and one likely explanation for this is higher diversity of sonotypes occupying those frequency bins (Figure S1). 75 Figure 3 - RMS (Root Mean Square) level of the PSD (Power Spectrum Density), percentiles and SPD (spectral probability density) of four points in the wind farm area (distances from the turbines as it shows). The control area is represented by Control 1 and 2. Red arrows indicate examples of peaks and katydid’s activity. 76 3.3. False colour spectrograms Figure 4 – Example of False Colour Spectrograms of the same day in the four points recorded. Indices used in this visualization were BGN (red), PMN (green), R3D (blue), calculated at a resolution of 30 seconds. FCS on the top represent the diel patterns, with 24h at a resolution of one minute every 15 minutes. At the bottom, a zoom in the morning built with one hour of continuous recordings. Distance between points 1 and 2, and 3 and 4 is 200 meters. Between 2 and 3 is 400 meters. The FCS shows (in red along the bottom of the plots) the amount of background noise registered through the gradient distance from the turbines (Figure 4). In the first point (100 meters from the turbines) this noise band is evident up to 4 kHz, but 900 m away from the turbines it is reduced to the first kHz. The insect activity appears dominating the night period in all four sample points, with apparently no evident differences. When we look at the bird activity, however, there are some clear differences between the points, as shown in the bottom FCS in Figure 4. Those differences can be seen in green and blue colours, in short 77 duration signals between 2 and 5 kHz. The long duration signals, represented as horizontal bands during the entire one-hour period, reflect Orthopteran activity. 3.4. Acoustic Indices The two indices chosen to inspect the variation in acoustic activity in a distance gradient presented similar patterns, with values increasing with distance, both in the morning (Figure 5; SNR: Chi-squared (1) = 13.65, p<0.001; ENT: Chi-squared (1) = 7.48, p<0.01) and in night periods (Figure 6; SNR: Chi-squared (1) = 22.14, p<0.001; ENT: Chi-squared (1) = 11.74, p<0.001). Results from model selections are presented in Table 2. Figure 5 – Acoustic indices variation during the morning period (from 0500 to 0800). On top, boxplots showing the distribution of Indices values, according to distance. W1 to W4 refers to the wind farm area, from the closest to the most distant point from the turbines. C1 and C2 refers to Control points, sampled in a different area. 78 Figure 6 – Acoustic indices variation during night periods recordings (from 1700 to 2000). On top, boxplot showing the distribution of Indices values, according to distance. W1 to W4 refers to the wind farm area, from the closest to most distant point from turbines. C1 and C2 refers to Control points. Table 2 – Results of mixed models selected from comparison with null models (ANOVA). All models include as random factors area (treatment/control) and time (hour and day, per sampling point). All models were checked for normality and homoscedasticity. Model Coefficients Period Index Random variables (SD) Fixed variable (SE) Day SNR Area: 0.022 (0.15) Distance: 0.130 (0.042) Time: 0.034 (0.184) Day ENT Area: 0.008 (0.09) Distance: 0.067 (0.035) Time: 0.033 (0.182) Night SNR Area: <0.001 Distance: 0.049 (0.007) Time: 0.032 (0.179) Night ENT Area: <0.001 Distance: 0.027 (0.007) Time: <0.001 79 4. Discussion We addressed the impacts of wind farms using three methods (FCS, SPD and two Acoustic Indices) to explore how the soundscapes dynamics are being affected by proximity to wind turbines. The frequency bins used by most birds overlapped with the noise emitted by the turbines, while only a few insects suffer this acoustic masking. Our results suggest that both birds and insects have different activity patterns associated with wind farm turbines, denouncing that the effect of wind farming on natural soundscapes is noticeable at small spatial scales. 4.1. Soundscape characterization Most calls from birds and insects (Figure 2) did not overlap in their dominant frequencies, although they do in time, with some insects also calling during the day. The Orthopterans dominated the night soundscape, with temporal partitioning in relation to other insect groups frequent in tropical regions - the day time active cicadas (Diwakar & Balakrishnan, 2007). The frequency range used by insects in our study (dominant frequencies between 3.1 kHz and 18.9 kHz) is similar to the one found in previous studies (Ferreira et al., 2018; Gasc et al., 2013; Penone et al., 2013; Phillips, 2018; Pijanowski et al., 2011), and the calling patterns we observed for Orthopterans differ between groups, with katydids using a broad frequency range and crickets calling with pure tone songs (Figure S1). Inasmuch, crickets seem to separate their signals better in the frequency spectrum (Schmidt & Balakrishnan, 2014) and this partitioning is also related to known patterns of highly diverse communities (Kostarakos et al., 2008; Schmidt et al., 2013). Birds were vocally active during the day, especially in the first hours of the morning – the dawn chorus. This phenomenon is widely studied, and theories about the reasons why it 80 occurs include optimization of foraging time and favourable time for sound transmission in the atmosphere (Berg et al., 2006; Hutchinson, 2002). Dominant frequency in bird calls varied from 515 Hz to 7.5 kHz, a range also comparable to other studies (Ferreira et al., 2018; Wiley & Richards, 1982) and potentially more affected by anthropogenic noise (Slabbekoorn & Ripmeester, 2008). The potential noise masking effect on the lower frequency bands (up to 4 kHz) could be visualized on False Colour Spectrograms (FCS), as a drastic reduction of bioacoustic signals on lower frequency bands dominated by noise (reddish colour, bottom left panel, Figure 4). Other possible explanation for the lack of signals in those frequency bands could be the displacement of animals that use those bands for reasons other than acoustic masking, such as avoidance of the physical structure of the wind farm. It is also important to remember that, even if the individuals are not directly affected by acoustic masking of their signals, they can still be impacted by the noise depending on the frequency range sensibility of auditory system (Robert & Beason, 2004). Concerning the community of animals in such affected areas, the acoustic niche hypothesis predicts that animals will compete for the acoustic space, occupying separate frequency bands (Krause, 1993). Hence, an environment richer in acoustic signals can also be considered more biodiverse. The Spectral Probability Density (SPD) plot summarizes the information about spectral values and shows that the sample points with more peaks in Root- mean-square values can be considered richer acoustically (more diversity in signal types), or having more complex acoustic communities. RMS of the raw audio signal was successfully used to track temporal and spatial dynamics both in terrestrial (Eldridge et al., 2018; Rodriguez et al., 2014) and marine environments (Bertucci et al., 2016). 81 4.2. Acoustic Indices Our results shown that acoustic activity increases with distance from the turbines, both during the morning and at night, suggesting an impact on bird and insect communities. These results are in contrast with a recent study in a temperate region (Florentin et al., 2015) that investigated soundscape dynamics in the vicinity of a single wind turbine. The authors found no impact on bird community based on ACI and Spectral Entropy indices. However, they also highlight the difficulty of making general assumptions due to limitations of the study such as low sampling rate, short time period and limited sampling (Florentin et al., 2015). We believe that our results were different because the area sampled by Florentin et al. (2015) only have one wind turbine and it is placed in a region with low bird diversity. Another study using acoustic indices to measure anthropic impacts on tropical areas showed results similar to ours. Duarte et al. (2015) showed significant differences in ACI (Acoustic Complexity Index) in an area with iron mining activity and a control area within the Brazilian Atlantic Forest, with higher biophony values away from the mining area. Most studies addressing anthropic impact on acoustic communities are focused on birds (Buxton et al., 2018), because they are good models on terrestrial environments: they are widespread, have their taxonomy and call identification well established, and are cost- efficient to monitor (Venier et al., 2012). Based on the spectral distribution of bird song (Figure 2), we found that the acoustic impact from turbine noise affects more the frequency bands used by birds, as documented in the literature (Slabbekoorn & Ripmeester, 2008). In fact, birds are considered the most impacted faunal group around wind farm facilities, according to previous studies (e.g. Dahl et al., 2012; Percival, 2005; Tellería, 2009). Here, even selecting only the first hours in the morning to sample specifically the avian dawn chorus, we 82 could still detect the presence of insects, but it is much less evident than at night. This represents an extra challenge in the use of acoustic indices to measure the specific activity of birds and could explain why some acoustic indices traditionally used for birds do not work (e.g. ACI, Pieretti et al., 2011). The abandonment of generalization of acoustic indices is an issue recently addressed by researchers. Particular indices may not be suitable for different regions, considering how variable the acoustic communities and the data sampling are (Bradfer‐Lawrence et al., 2019; Sugai et al., 2019). We observed diminished Orthopteran calling activity close to the wind turbines, an effect that could be explained by acoustic masking. Insects are an interesting group to be used in environmental monitoring (Riede, 1998), although it is under investigated, mainly due to lack of baseline studies. But it is known that strategies to avoid acoustic masking also reported for vertebrates can occur in invertebrates as well. This includes increasing the amplitude of the calls (Duarte et al., 2019; Nemeth & Brumm, 2010), spatially moving away from noise source (Bunkley et al., 2017), or increase the intensity of masking frequencies (Lampe et al., 2012). Other strategies common in vertebrates, such as changing calling time are not common (Raboin & Elias, 2019) but can occur in invertebrates (Duarte et al., 2019). Further explorations could focus on species identification and measurement of call parameters to better understand the phenomenon behind the decrease observed. 5. Conclusions The eolic energy industry is an interesting disturbance model, as the facilities can be placed in areas with less interference in the vegetation than other energy plants, such as hydroelectric. Thus, apart from direct collision (Bispo et al., 2013; Grodsky et al., 2013), other 83 changes in animal community are related to the acoustic impact from the turbines noise in these areas (Rabin et al., 2006; Zwart et al., 2016, this study). Recently the use of acoustic monitoring as an environmental assessment tool has been supported and encouraged. In addition to other traditional methods (Burivalova et al., 2019), the use of acoustic indices can measure and quantify human-driven changes in the environment (Doser et al., 2019). Here, we found that: (1) the acoustic diversity, represented by occupation of frequency bands, increases with distance from the turbines; (2) both ENT and SNR indices capture the acoustic variation along a distance gradient, during morning and evening periods; (3) birds that call in lower frequencies suffer more from acoustic overlap than insects. We suggest that further studies should employ multiple acoustic indices on modelling the anthropic disturbance on soundscapes, once the combination of acoustic indices can provide a better understanding of soundscape patterns (Eldridge et al., 2018). We also recommend that communities of soniferous insects are studied, filling the gap of information that exists for tropical insects, and exploring the use of this group in environmental health measurements (Duarte et al., 2019; Raboin & Elias, 2019). The use of soundscape-based analyses to inform and guide environmental policies is still incipient. In the case of wind farms, we recommend the inclusion of noise propagation modelling taking into account also the vocal characteristics of the animals in the area, as well as the spatial configuration of the wind farms. The size of the wind farms and the proximity of turbines to each other may have unexpected local effects on the range of area impacted by their noise. As shown here, species can be impacted in a different way. Therefore, the use of soundscape-based analyses in environmental monitoring should be employed with 84 caution, since impacts will depend on local propagation, spatial and temporal configuration of human activities and the features of local animal communities. 6. References Aide, T. M., Hernández-Serna, A., Campos-Cerqueira, M., Acevedo-Charry, O., & Deichmann, J. L. (2017). Species richness (of insects) drives the use of acoustic space in the tropics. Remote Sensing, 9(11). https://doi.org/10.3390/rs9111096 Anderson, R., Morrison, M., Sinclair, K., & Strickland, D. (1999). Studying wind energy/bird interactions: A guidance document, (December), 1–94. Barber, J. R., Crooks, K. R., & Fristrup, K. M. (2010). The costs of chronic noise exposure for terrestrial organisms. Trends in Ecology and Evolution, 25(3), 180–189. https://doi.org/10.1016/j.tree.2009.08.002 Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models Using {lme4}. Journal of Statistical Software, 67(1), 1–48. https://doi.org/10.18637/jss.v067.i01 Berg, K. S., Brumfield, R. T., & Apanius, V. (2006). Phylogenetic and ecological determinants of the neotropical dawn chorus. Proceedings of the Royal Society B: Biological Sciences, 273(1589), 999–1005. https://doi.org/10.1098/rspb.2005.3410 Bertucci, F., Parmentier, E., Lecellier, G., Hawkins, A. D., & Lecchini, D. (2016). Acoustic indices provide information on the status of coral reefs: An example from Moorea Island in the South Pacific. Scientific Reports, 6(February), 1–9. https://doi.org/10.1038/srep33326 Bispo, R., Bernardino, J., Marques, T. A., & Pestana, D. (2013). Modeling carcass removal time for avian mortality assessment in wind farms using survival analysis. Environmental and Ecological Statistics, 20(1), 147–165. https://doi.org/10.1007/s10651-012-0212-5 Bradbury, J. W., & Vehrencamp, S. L. (1998). Principles of Animal Communication (1st ed.). Sinauer Associates Inc. Bradfer‐Lawrence, T., Gardner, N., Bunnefeld, L., Bunnefeld, N., Willis, S. G., & Dent, D. H. (2019). Guidelines for the use of acoustic indices in environmental research. Methods in Ecology and Evolution, 2019(June), 2041-210X.13254. https://doi.org/10.1111/2041- 210X.13254 85 Bunkley, J. P., McClure, C. J. W., Kawahara, A. Y., Francis, C. D., & Barber, J. R. (2017). Anthropogenic noise changes arthropod abundances. Ecology and Evolution, 7(9), 2977– 2985. https://doi.org/10.1002/ece3.2698 Burivalova, Z., Allnutt, T. F., Rademacher, D., Schlemm, A., Wilcove, D. S., & Butler, R. A. (2019). What works in tropical forest conservation, and what does not: Effectiveness of four strategies in terms of environmental, social, and economic outcomes. Conservation Science and Practice, (March), e28. https://doi.org/10.1111/csp2.28 Burivalova, Z., Game, E. T., & Butler, R. A. (2019). The sound of a tropical forest. Science, 363(6422), 28–29. https://doi.org/10.1126/science.aav1902 Buxton, R. T., Agnihotri, S., Robin, V. V, Goel, A., Balakrishnan, R., State, C., Pradesh, A. (2018). Acoustic indices as rapid indicators of avian diversity in different land-use types in an Indian biodiversity hotspot. Journal of Ecoacoustics, 2, 1–17. Dahl, E. L., Bevanger, K., Nygård, T., Røskaft, E., & Stokke, B. G. (2012). Reduced breeding success in white-tailed eagles at Smøla windfarm, western Norway, is caused by mortality and displacement. Biological Conservation, 145(1), 79–85. https://doi.org/10.1016/j.biocon.2011.10.012 Darras, K., Pütz, P., Fahrurrozi, Rembold, K., & Tscharntke, T. (2016). Measuring sound detection spaces for acoustic animal sampling and monitoring. Biological Conservation, 201(September), 29–37. https://doi.org/10.1016/j.biocon.2016.06.021 de Lucas, M., Ferrer, M., Bechard, M. J., & Muñoz, A. R. (2012). Griffon vulture mortality at wind farms in southern Spain: Distribution of fatalities and active mitigation measures. Biological Conservation, 147(1), 184–189. https://doi.org/10.1016/j.biocon.2011.12.029 Depraetere, M., Pavoine, S., Jiguet, F., Gasc, A., Duvail, S., & Sueur, J. (2012). Monitoring animal diversity using acoustic indices: Implementation in a temperate woodland. Ecological Indicators, 13(1), 46–54. https://doi.org/10.1016/j.ecolind.2011.05.006 Diwakar, S., & Balakrishnan, R. (2007). The assemblage of acoustically communicating crickets of a tropical evergreen forest in southern india: Call diversity and diel calling patterns. Bioacoustics, 16(2), 113–135. https://doi.org/10.1080/09524622.2007.9753571 Doser, J. W., Finley, A. O., Kasten, E. P., & Gage, S. H. (2019). Assessing soundscape disturbance through hierarchical models and acoustic indices: a case study on a shelterwood logged northern Michigan forest. Retrieved from http://arxiv.org/abs/1911.03278 Drake, D., Jennelle, C. S., Liu, J.-N., Grodsky, S. M., Schumacher, S., & Sponsler, M. (2015). Regional Analysis of Wind Turbine-Caused Bat Mortality. Acta Chiropterologica, 17(1), 179–188. https://doi.org/10.3161/15081109ACC2015.17.1.015 86 Duarte, M., Caliari, E. P., Scarpelli, M. D. A., Lobregat, G. O., Young, R. J., & Sousa-Lima, R. S. (2019). Effects of mining truck traffic on cricket calling activity. The Journal of the Acoustical Society of America, 146(1), 656–664. https://doi.org/10.1121/1.5119125 Duarte, M. H. L., Sousa-Lima, R. S., Young, R. J., Farina, A., Vasconcelos, M., Rodrigues, M., & Pieretti, N. (2015). The impact of noise from open-cast mining on Atlantic forest biophony. Biological Conservation, 191(August), 623–631. https://doi.org/10.1016/j.biocon.2015.08.006 Eldridge, A., Guyot, P., Moscoso, P., Johnston, A., Eyre-Walker, Y., & Peck, M. (2018). Sounding out ecoacoustic metrics: Avian species richness is predicted by acoustic indices in temperate but not tropical habitats. Ecological Indicators, 95(September), 939–952. https://doi.org/10.1016/j.ecolind.2018.06.012 Farina, A. (2014). Soundscape Ecology. https://doi.org/10.1007/978-94-007-7374-5 Ferreira, L. M., Oliveira, E. G., Lopes, L. C., Brito, M. R., Baumgarten, J., Rodrigues, F. H., & Sousa-Lima, R. S. (2018). What do insects, anurans, birds, and mammals have to say about soundscape indices in a tropical savanna. Journal of Ecoacoustics, 2(March), PVH6YZ. https://doi.org/10.22261/JEA.PVH6YZ Florentin, J., Fauville, B., Gérard, M., Moiny, F., Rasmont, P., Kouroussis, G., & Verlinden, O. (2015). Soundscape Analysis and Wildlife Presence in the Vicinity of a Wind Turbine. EuroNoise 2015, (c), 2689–2694. Fuller, S., Axel, A. C., Tucker, D., & Gage, S. H. (2015). Connecting soundscape to landscape: Which acoustic index best describes landscape configuration? Ecological Indicators, 58(January 2016), 207–215. https://doi.org/10.1016/j.ecolind.2015.05.057 Gasc, A., Sueur, J., Pavoine, S., Pellens, R., & Grandcolas, P. (2013). Biodiversity Sampling Using a Global Acoustic Approach: Contrasting Sites with Microendemics in New Caledonia. PLoS ONE, 8(5). https://doi.org/10.1371/journal.pone.0065311 Grodsky, S. M., Jennelle, C. S., & Drake, D. (2013). Bird Mortality at a Wind-Energy Facility near a Wetland of International Importance. The Condor, 115(4), 700–711. https://doi.org/10.1525/cond.2013.120167 Hutchinson, J. M. C. (2002). Two explanations of the dawn chorus compared: How monotonically changing light levels favour a short break from singing. Animal Behaviour, 64(4), 527–539. https://doi.org/10.1006/anbe.2002.3091 Kostarakos, K., Hartbauer, M., & Ro, H. (2008). Matched Filters , Mate Choice and the Evolution of Sexually Selected Traits, 3(8), 16–18. https://doi.org/10.1371/journal.pone.0003005 87 Krause, B. L. (1993). The Niche Hypothesis: A Virtual Symphony of Animal Sounds, The Origins of Musical Expression and the Health of Habitats. The Soundscape Newsletter, (6), 6– 10. Retrieved from http://interact.uoregon.edu/Medialit/wfae/library/newsletter/SNL6.PDF Kunc, H. P., & Schmidt, R. (2019). The effects of anthropogenic noise on animals: a meta- analysis. Biology Letters, 15, 5. https://doi.org/10.1093/beheco/aru105 Kunz, T. H., Arnett, E. B., Erickson, W. P., Hoar, A. R., Johnson, G. D., Larkin, R. P., … Turtle, M. D. (2007). Ecological impacts of wind energy development on bats: Questions, research needs, and hypotheses. Frontiers in Ecology and the Environment, 5(6), 315–324. https://doi.org/10.1890/1540-9295(2007)5[315:EIOWED]2.0.CO;2 Lampe, U., Schmoll, T., Franzke, A., & Reinhold, K. (2012). Staying tuned : grasshoppers from noisy roadside habitats produce courtship signals with elevated frequency components, 1348–1354. https://doi.org/10.1111/1365-2435.12000 Merchant, N. D., Fristrup, K. M., Johnson, M. P., Tyack, P. L., Witt, M. J., Blondel, P., & Parks, S. E. (2015). Measuring acoustic habitats. Methods in Ecology and Evolution, 6(3), 257– 265. https://doi.org/10.1111/2041-210X.12330 Merchant, N. D., Thompson, P. M., Dorocicz, J., Pirotta, E., Barton, T. R., & Dakin, D. T. (2013). Spectral probability density as a tool for marine ambient noise analysis. The Journal of the Acoustical Society of America, 133(5), 3494–3494. https://doi.org/10.1121/1.4806195 Nemeth, E., & Brumm, H. (2010). Birds and Anthropogenic Noise : Are Urban Songs Adaptive?, 176(4). https://doi.org/10.1086/656275 Parks, S. E., Searby, A., Célérier, A., Johnson, M. P., Nowacek, D. P., & Tyack, P. L. (2011). Sound production behavior of individual North Atlantic right whales: Implications for passive acoustic monitoring. Endangered Species Research, 15(1), 63–76. https://doi.org/10.3354/esr00368 Patricelli, G. L., & Blickley, J. L. (2006). Avian Communication in Urban Noise: Causes and Consequences of Vocal Adjustment. The Auk, 123(3), 639. https://doi.org/10.1642/0004- 8038 Penone, C., Le Viol, I., Pellissier, V., Julien, J.-F., Bas, Y., & Kerbiriou, C. (2013). Use of Large-Scale Acoustic Monitoring to Assess Anthropogenic Pressures on Orthoptera Communities. Conservation Biology, 27(5), 979–987. https://doi.org/10.1111/cobi.12083 Percival, S. M. (2005). Birds and windfarms: What are the real issues? British Birds, 98(4), 194–204. Retrieved from https://www.scopus.com/inward/record.uri?eid=2-s2.0- 23844443477&partnerID=40&md5=2c0734df52bfe8f458c122a93c78403c Phillips, Y. (2018). THESIS: Analysis and Visualisation of very-long-duration acoustic recordings of the natural environment. 88 Pieretti, N., Duarte, M. H. L., Sousa-Lima, R. ., Rodrigues, M., Young, R. J., & Farina, A. (2015). Determining temporal sampling schemes for passive acoustic studies in different tropical ecosystems. Tropical Conservation Science, 8(1), 215–234. Pieretti, N., Farina, A., & Morri, D. (2011). A new methodology to infer the singing activity of an avian community: The Acoustic Complexity Index (ACI). Ecological Indicators, 11(3), 868–873. https://doi.org/10.1016/j.ecolind.2010.11.005 Pijanowski, B. C., Villanueva-Rivera, L. J., Dumyahn, S. L., Farina, A., Krause, B. L., Napoletano, B. M., … Pieretti, N. (2011). Soundscape Ecology: The Science of Sound in the Landscape. BioScience, 61(3), 203–216. https://doi.org/10.1525/bio.2011.61.3.6 R Core Team. (2018). R: A language and environment for statistical computing. Vienna, Austria.: R Foundation for Statistical Computing. Rabin, L. A., Coss, R. G., & Owings, D. H. (2006). The effects of wind turbines on antipredator behavior in California ground squirrels (Spermophilus beecheyi). Biological Conservation, 131(3), 410–420. https://doi.org/10.1016/j.biocon.2006.02.016 Raboin, M., & Elias, D. O. (2019). Anthropogenic noise and the bioacoustics of terrestrial invertebrates. The Journal of Experimental Biology, 222(12), jeb178749. https://doi.org/10.1242/jeb.178749 Reed, S. E., Boggs, J. L., & Mann, J. P. (2012). A GIS tool for modeling anthropogenic noise propagation in natural ecosystems. Environmental Modelling and Software, 37, 1–5. https://doi.org/10.1016/j.envsoft.2012.04.012 Riede, K. (1998). Acoustic monitoring of Orthoptera and its potential for conservation. Journal of Insect Conservation, 2(3–4), 217–223. https://doi.org/10.1023/A:1009695813606 Robert, C., & Beason, R. C. (2004). What Can Birds Hear ? UC Agriculture & Natural Resources Proceedings of the Vertebrate Pest Conference, 21, 92–96. Rodriguez, A., Gasc, A., Pavoine, S., Grandcolas, P., Gaucher, P., & Sueur, J. (2014). Temporal and spatial variability of animal sound within a neotropical forest. Ecological Informatics, 21, 133–143. https://doi.org/10.1016/j.ecoinf.2013.12.006 Schmidt, A. K. D., & Balakrishnan, R. (2014). Ecology of acoustic signalling and the problem of masking interference in insects. Journal of Comparative Physiology A: Neuroethology, Sensory, Neural, and Behavioral Physiology, 201(1), 133–142. https://doi.org/10.1007/s00359-014-0955-6 Schmidt, A. K. D., Romer, H., & Riede, K. (2013). Spectral niche segregation and community organization in a tropical cricket assemblage. Behavioral Ecology, 24(2), 470–480. https://doi.org/10.1093/beheco/ars187 89 Shannon, G., McKenna, M. F., Angeloni, L. M., Crooks, K. R., Fristrup, K. M., Brown, E., … Wittemyer, G. (2016). A synthesis of two decades of research documenting the effects of noise on wildlife. Biological Reviews, 91(4), 982–1005. https://doi.org/10.1111/brv.12207 Slabbekoorn, H., & den Boer-Visser, A. (2006). Cities Change the Songs of Birds. Current Biology, 16(23), 2326–2331. https://doi.org/10.1016/j.cub.2006.10.008 Slabbekoorn, H., & Ripmeester, E. A. P. (2008). Birdsong and anthropogenic noise: Implications and applications for conservation. Molecular Ecology, 17(1), 72–83. https://doi.org/10.1111/j.1365-294X.2007.03487.x Smallwood, K. S., Rugge, L., & Morrison, M. L. (2009). Influence of behavior on bird mortality in wind energy developments. Journal of Wildlife Management, 73(7), 1082–1098. https://doi.org/10.2193/2008-555 Sugai, L. S., Llusia, D., Sanna, T., Silva, F., & Desjonqu, C. (2019). A roadmap for survey designs in terrestrial acoustic monitoring, 1–16. https://doi.org/10.1002/rse2.131 Tellería, J. L. (2009). Potential impacts of wind farms on migratory birds crossing Spain. Bird Conservation International, 19(2), 131–136. https://doi.org/10.1017/S0959270908008137 Towsey, M., Truskinger, A., & Roe, P. (2016). Audio Analysis Software V. 18.6.3.3. Brisbane. Retrieved from https://github.com/QutEcoacoustics/audio- analysis Towsey, Michael. (2017). Title: The calculation of acoustic indices derived from long- duration recordings of the natural environment. Retrieved from http://eprints.qut.edu.au/61399/. Towsey, Michael, Zhang, L., Cottman-Fields, M., Wimmer, J., Zhang, J., & Roe, P. (2014). Visualization of long-duration acoustic recordings of the environment. Procedia Computer Science, 29, 703–712. https://doi.org/10.1016/j.procs.2014.05.063 Tucker, D., Gage, S. H., Williamson, I., & Fuller, S. (2014). Linking ecological condition and the soundscape in fragmented Australian forests. Landscape Ecology, 29(4), 745–758. https://doi.org/10.1007/s10980-014-0015-1 Venier, L. A., Holmes, S. B., Holborn, G. W., McIlwrick, K. A., & Brown, G. (2012). Evaluation of an automated recording device for monitoring forest birds. Wildlife Society Bulletin, 36(1), 30–39. https://doi.org/10.1002/wsb.88 Wiley, R. H., & Richards, D. G. (1982). Adaptations for Acoustic Communication in Birds: Sound Transmission and Signal Detection. Acoustic Communication in Birds (Vol. 1). https://doi.org/10.1016/b978-0-08-092416-8.50014-0 90 Wrightson, K. (1999). An introduction to acoustic ecology. Journal of Electroacoustic Music, 12, 10–13. Retrieved from http://nova.nefsc.noaa.gov/palit/Wrightson_2000_The Journal of Acoustic Ecology_An Introduction to Acoustic Ecology.pdf Zwart, M. C., Dunn, J. C., McGowan, P. J. K., & Whittingham, M. J. (2016a). Wind farm noise suppresses territorial defense behavior in a songbird. Behavioral Ecology, 27(1), 101– 108. https://doi.org/10.1093/beheco/arv128 91 Supplementary material Table S1 – Bird species identified from a 300 minutes sample in the wind farm area. Taxon English Name Tinamidae Crypturellus parvirostris Small-billed Tinamou Accipitridae Rupornis magnirostris Roadside Hawk Charadriidae Vanellus chilensis Southern Lapwing Columbidae Columbina picui Picui Ground-Dove Strigidae Glaucidium brasilianum Ferruginous Pygmy-Owl Bucconidae Nystalus maculatus Spot-backed Puffbird Picidae Veniliornis passerinus Little Woodpecker Cariamidae Cariama cristata Red-legged Seriema Falconidae Caracara plancus Southern Caracara Thamnophilidae Myrmorchilus strigilatus Stripe-backed Antbird Formicivora melanogaster Black-bellied Antwren Dendrocolaptidae Lepidocolaptes angustirostris Narrow-billed Woodcreeper Rhynchocyclidae Tolmomyias flaviventris Yellow-breasted Flycatcher Todirostrum cinereum Common Tody-Flycatcher Hemitriccus margaritaceiventer Pearly-vented Tody-tyrant Tyrannidae Euscarthmus meloryphus Tawny-crowned Pygmy-Tyrant Phaeomyias murina Mouse-colored Tyrannulet Myiarchus tyrannulus Brown-crested Flycatcher Pitangus sulphuratus Great Kiskadee Empidonomus varius Variegated Flycatcher Vireonidae Cyclarhis gujanensis Rufous-browed Peppershrike Troglodytidae Troglodytes musculus Southern House Wren Cantorchilus longirostris Long-billed Wren Polioptilidae 92 Polioptila plumbea Tropical Gnatcatcher Icteridae Icterus jamacaii Campo Troupial Thraupidae Coryphospingus pileatus Pileated Finch Coereba flaveola Bananaquit Figure S1 – Spectrograms (FFT=512) of 30 seconds of sounds recorded at 19h15 in two different areas in Rio Grande do Norte state in Brazil. Above the wind farm area and below the control area. Patterns observed (dark banding) represent Orthopteran acoustic activity. On wind farm area, the arrow points at katidids’ acoustic activity. 93 94 Acoustic Indices predict bird species richness and community composition in tropical biomes Eliziane G. Oliveiraa,c, Lucas Gasparb, Carlos Gussonid, Vinicius Rodrigues Tonettib, Alan Glaucoa, Jorge Dantasa, Luane M. Ferreiraa, Milton Ribeirob, Renata Sousa-Limaa a Laboratório de Bioacústica (LaB), Departamento de Fisiologia e Comportamento, Biosciences Center, Universidade Federal do Rio Grande do Norte, Avenida Senador Salgado Filho, 3000, Bairro Lagoa Nova, Natal, RN. 59078-970, Brazil. b Spatial Ecology and Conservation lab (LEEC), Department of Ecology, State University of Sao Paulo (UNESP), Av. 24 A, 1515, Rio Claro SP, 13506-900, Brazil. c Graduate Program in Ecology, Biosciences Center, Universidade Federal do Rio Grande do Norte, Avenida Senador Salgado Filho, 3000, Bairro Lagoa Nova, Natal, RN. 59078-970, Brazil. d Rua 12-B, 621, Vila Indaiá, Rio Claro, SP. CEP: 13506-746 95 96 Abstract Passive acoustic monitoring is considered an important tool to quantify biodiversity and track changes in animal acoustic activity. To analyse the large volume of recordings, acoustic indices have been developed to summarize the information in the recordings. However, the relationship between indices developed to quantify animal activity and the traditional estimates of biodiversity is still controversial. Here we tested whether single indices can predict bird species richness and if a composition of several indices allow discrimination of different species’ composition in three biomes in Brazil: Cerrado (Savanna), Caatinga (Seasonally Dry Forest), and Atlantic Forest (Tropical Moist Broadleaf Forest). We identified a total of 132 bird species: 25 in the Cerrado, 53 in the Caatinga, and 69 in the Atlantic Forest. We found that for all biomes pooled together, NDSI was correlated to the number of bird species identified in each recording. Other indices were correlated when considering only Cerrado (NDSI) and Atlantic Forest (negative correlation with AR). In Caatinga no single index was correlated with bird diversity. The reasons why those indices perform differently are discussed based on the spectral characteristics of the birds’ calls and species’ predominance. The combination of six acoustic indices allowed discrimination of these three biomes. Our results indicate that the use of acoustic indices as a proxy of avian diversity in tropical environments can vary, according to characteristics of the avian community and phytophysignomies. Future studies should better explore how those local peculiarities can be considered in calculations and interpretation of results. Keywords: Ecoacoustic indices, Soundscape, Biodiversity Monitoring, Avian Community 97 1. Introduction Passive acoustic monitoring has become popular in the last decade as an important tool to measure ecological processes, temporal and spatial changes in animal activity patterns (Blumstein et al., 2011; Campos-Cerqueira & Aide, 2016; Lellouch et al., 2014; Machado et al., 2017), and to identify the richness of determined group (Shonfield & Bayne, 2017; Wimmer et al., 2013; Zhang et al., 2016). Birds are one of the most studied group in terrestrial passive acoustic monitoring studies (Sugai et al., 2019), which enables sampling distinct areas simultaneously for long periods of time, with no interruption. It can help, for example, the detection of rare species which would go unnoticed in a bird survey using traditional methodology, such as fixed-radius sampling points and mist nests (Holmes et al., 2014). Recent studies showed that the acoustic monitoring for counting species is not only a complimentary methodology, but can be used alone, as the results are similar to other traditional methodologies (Alquezar & Machado, 2015), or can even outperform direct observations by humans (Darras et al., 2019). The major challenge in using acoustic monitoring is the large amount of data generated, which makes it difficult for experts to hear all the files (Towsey, Wimmer, Williamson, & Roe, 2014). For this reason, several recording and analysis schemes have been tested, so that the manual effort of the specialists can be optimized. One efficient scheme to sample diurnal birds is the analysis of dawn chorus, when animals are more active (Wimmer et al., 2013). With the premise that richer communities are expected to have richer acoustic environments, ecological indices that unveil this acoustic diversity can be used as a proxy of biodiversity (Sueur et al., 2014). In the past years computational tools were developed in order to improve the use of increasingly larger acoustic datasets in environmental monitoring, behavioral and ecological studies, especially through acoustic indices (Burivalova et al., 2017; 98 Towsey et al., 2018; Zhang et al., 2016). Acoustic indices were introduced in the field of acoustic ecology to summarize information within acoustic recording, allowing researchers inferring processes occurring in a landscape in an automated way (Sueur et al., 2014). Previous work comparing acoustic indices with other methodologies of estimating bird diversity and richness were carried on a variety of environments, and sampling designs and the findings about which index (or indices combinations) better summarize soundscape information are variable (Table 1). Factors that may be causing this variation are related to birds’ vocal features, such as the shape of sound units and their amplitude, and also to the frequency of occurrence of each species within communities (Zhao et al., 2019). Vocal features can be related to the vegetation where the species vocalize. Areas with a dense vegetation will attenuate higher frequency sounds, while in areas with open vegetation, the sound tends to propagate more freely allowing all frequencies to be detected at distance (Bradbury & Vehrencamp, 1998). Such peculiarities may hinder the generalization of findings so studies testing the applicability of using sounds to characterize and monitor biodiversity should consider the effect that different habitats have on the results. 99 Table 1 – Summary of studies that investigated the relationship between acoustic indices and animal species’ richness. Reference Spatial Area Sampling Species Acoustic Indices Main results design design richness estimative Depraetere Three Forest- 74 days, 3 hrs Manual Acoustic Richness, Dissimilarity Acoustic richness in agreement with et al., 2012 recorders, cropland dawn, 3 hrs inspection Index traditional aural identification. Higher 300 m France dusk, 150s of values in young forest. Peak at dawn apart. every 15 min. subsample chorus. Dissimilarity Index indicated Gradient Total= 222 hrs differences between areas. of tree of recordings density at 44.1 kHz Mammides 97 Rubber Recordings Point Count Acoustic Richness, Acoustic No index with strong correlation. Total et al., 2017 recorders, plantations, made at the surveys Complexity, Biodiversity Index, Entropy, Acoustic diversity, Acoustic 250m tropical same time the NDSI, Acoustic Diversity, Acoustic evenness moderate correlations. apart. and count surveys, Evenness Environmental dissimilarity didn't affect Two subtropical for the same indices performance different forest; duration (15 areas China. min) at 44.1 kHz. Towsey et One Open 5 days, Manual Average Signal Amplitude, Spectral diversity*, Spectral and Temporal al., 2014 recorder forest, with continuous inspection Background Noise, Signal to Entropy, Acoustic complexity were the best Eucalyptus. recordings at files Noise Ratio, Acoustic activity, indicators. Australia 22.05 kHz Count of acoustic events, Average duration of acoustic *best index in single performance events, Entropy of the signal envelope, Acoustic Complexity, Mid-band activity, Entropy of the average spectrum, Entropy of spectral maxima, Entropy of spectral variance, Spectral diversity, Spectral persistence. 100 Reference Spatial Area Sampling Species Acoustic Indices Main results design design richness estimative Izaguirre, 60 points Dry Continuous Bird point Acoustic evenness, Acoustic Acoustic complexity correlated to bird et al., 2018 (5 Tropical recordings (2 counts. diversity, Acoustic complexity, abundance, Number of Peaks with bird recorders Forest. hr) during Bird Biodiversity index, NDSI, Total diversity. in each of Costa Rica dawn and abundance, Entropy, Median Amplitude 12 areas) dusk, 10 min richness, Envelope, Number of peaks 200 m every hour for diversity apart the rest of the and day. 2 evenness consecutive days. Sampling rate of 44.1 kHz. Machado et 30 Points Savannah. 15 minutes in Manual Acoustic diversity Index, NDSI Acoustic Diversity Index associated with al., 2017 Brazil each point. inspection species richness, values of NDSI lower in Sampling rate of the files areas close to highways of 48 kHz. Moreno- 3 areas Valdivian 1 min per Manual Spectral Entropy, Temporal Temporal Entropy, Acoustic Diversity Index Gómez et (600 m Forest. hour, inspection Entropy, Total Entropy, Acoustic and Acoustic evenness Index associated al., 2019 apart) Chile 24h/day. of the files Complexity Index, Acoustic with bird richness Sampling rate Diversity Index, Acoustic of 44.1 kHz. Evenness Index, Bioacoustic Index Jorge et al., 12 points Atlantic 10 minutes Point-count Total Entropy, Acoustic Moderate correlation with Acoustic 2018 Forest. recording/10 survey Complexity Index, Acoustic Evenness Index. Other correlations were Brazil minutes Diversity Index, Acoustic considered biased by the observer during paused. Evenness Index, Bioacoustic the data collection. Sampling rate Index, NDSI of 48 kHz. 101 Here, we used data from three tropical biomes in Brazil. A Seasonally Dry Tropical Forest (SDTF) in the Northeast referred as Caatinga, a Savanna in central Brazil, known as Cerrado, and the Tropical Moist Broadleaf Forest, the Brazilian Atlantic Forest on the Southeast region of the country. Our main goal was to understand how acoustic indices can be used to infer information about richness and species composition of bird communities, aiming to answer: (1) Can acoustic indices be used as a proxy of species richness considering different environments? (2) Can the features of birds’ calls influence the relationship between species richness and acoustic indices? And, (3) Can the differences in avian community composition among areas be predicted by acoustic indices? We hypothesize that (1) Indices can respond in two different ways to bird diversity: i) Indices that are based on occupation of frequency bands (e.g. Acoustic Diversity Index) will be correlated to species richness and ii) Indices based on measurement of activity (e.g. Total Entropy) will be related to the amount of acoustic activity, which may or may not have a relationship with richness and can indicate other processes occurring in the area, such as effects of human activities; (2) Indices will be affected by the frequency bandwidth occupied by birds’ calls. In areas where species present more even distribution of energy in frequency bands, indices based on occupation of those bands will present higher values; (3) Combination of acoustic indices will account for different features of the acoustic space, enabling the identification of differences among areas. 2. Methodology 2.1. Acoustic recordings 102 In all areas, we installed recorders during the rainy season, and the sampling minutes were sorted within a one-month period. In the selected month, we sampled one hour during the morning chorus, the period of time when most bird species are vocalizing (Wimmer et al., 2013). We used Song Meter Digital Field Recorders (SM2, in Cerrado. SM3 in Atlantic Forest and Caatinga) (Wildlife Acoustics, Inc., Massachusetts), positioned 1.5 m high, with at least 400 meters between them to avoid superposition (Figure 1). Sampling rate varied from 44.1 kHz to 96 kHz, 16 bit WAV format, but for the calculation of acoustic indices (see below) files were downsampled to 24 kHz. Other details about data collection are presented below for each site. 2.2. Study sites Caatinga is a Seasonally Dry Tropical Forest in northeastern Brazil. It is a semi-arid region, with precipitation varying around 200-800 mm annually, and a rainy season of 3 to 5 months (Olmos et al., 2005). Our study area, the Serra do Feiticeiro is considered one of the most important areas of Caatinga within the state of Rio Grande do Norte (Marinho et al., 2018). The area is not part of a legal reserve, and human occupation in small farms is present in the region. The area is composed of an open arbustive and arboreal vegetation, and most tree species do not pass 5m of height. We recorded at the end of rainy season, from June to July 2017. The Brazilian Savanna, known as Cerrado, is considered a biodiversity hotspot (Strassburg et al., 2017), with phytophysiognomies that range from open grassland to closed canopy forests and gallery forests adjacent water courses (Oliveira-Filho & Ratter, 2002). The climate is seasonal tropical, and the rainy season ranges from October to March (Queirolo & Motta-Junior, 2007). We collected data from November to December 2017, in the Serra da 103 Canastra National Park, Minas Gerais State. Vegetation in the areas was mainly composed by shrubs, small trees and open grassland. The Atlantic Forest is also considered a biodiversity hotspot (Myers et al., 2000) and covers areas along the coastline from southern to northeastern Brazil. The weather is usually warm and wet year round, but some phytophysiognomies experience a dry season from April to September (Morellato & Haddad, 2016). We recorded during November 2016, in a fragmented landscape area, and recorders were placed in areas at least 60m away from the border. Figure 1 – Nine recording sites distributed equally in three Tropical Biomes. Brazilian Savanna (Cerrado), Seasonally Dry Tropical Forest (Caatinga) and Tropical Broadleaf Forest (Atlantic Forest) were used to sample the soundscapes using autonomous recorders. Note the difference in spatial scale among the areas sampled. 2.3. Data analysis 104 For species identification, we subsampled our data by randomly sorting 60 second audio clips. For each biome we used 300 minutes, 100 in each of three points. Those were analysed by two ornithologists in each biome (AG and JD in Caatinga, VT and LG in Atlantic Forest, EO and CG in Cerrado), that identified the species aurally and visually inspecting the spectrograms, using the software Raven Pro 1.5. Annotations generated in the software were translated to a presence/absence matrix in R platform (Supplementary Material 1). From this data base we also investigated the ten species that were most identified in each biome, and also inspected the frequency range used by them. 2.4. Acoustic Indices To calculate the acoustic indices, we used data from one hour per day, for 25 days, for each of the nine points. In total, approximately 16200 one-minute files were processed. We chose six ecological indices available at R software (R Core Team, 2018) packages ‘soundecology’ (Villanueva-Rivera & Pijanowski, 2016) and ‘seewave’: Acoustic Diversity Index (ADI), Total Entropy (H), Acoustic Complexity Index (ACI), Bioacoustic Index (BI), Acoustic Richness (AR), Normalized Difference Soundscape Index (NDSI) (Table 2). Indices were calculated at a resolution of 60 seconds and maximum frequency values were set to 12 kHz, ensuring that bird calls were captured, but other sound sources occupying higher frequency were left out. Table 2 – Description of six Acoustic Indices used. Details about index calculation can be found on references. Index Description Reference ADI Acoustic Diversity Index. Is the result of applying the (Villanueva- Shannon Diversity Index to each frequency bin (the Rivera et al., default is 1kHz). Also, by default, there is a cutoff value of 2011) -50 dBFS, to remove the faint sounds. According to this index, greater values will represent greater number of 105 active bands and, therefore, more diverse acoustic communities. H Total Entropy. Is the product of both temporal and (Sueur et al., spectral entropies. Values are close to 0 for a single pure 2008) tone and increases up to 1 with the occupation of more frequency bands and amplitude modulations. ACI The Acoustic Complexity Index was designed to measure (Pieretti, Farina, biophony activity (mainly birds), considering the & Morri, 2011) variability in the amplitude values (greater than 3db) of close time units in each frequency bin. By using this calculation, it assumes that human-made sounds, usually constant and concentrated in lower frequency bins, are excluded BIO Bioacoustic Index was proposed to measure relative avian (Boelman et al., song abundance by calculating the area under the 2007) normalized power spectrum that includes all frequency bands and is greater than the minimum intensity of each curve. Minimum frequency was set to 500 Hz. NDSI Normalized Difference Soundscape Index, estimates the (Kasten et al., proportion of anthropogenic acoustic activity compared 2012) to biophonic in a soundscape. The anthropogenic frequency bin was defined from 0 to 500 Hz. AR Acoustic Richness. Is a ranked index based on the rank of (Depraetere et two other indices: M (Median of the amplitude envelope) al., 2012) and Ht (Temporal Entropy). 2.5. Statistical Analysis Aiming to compare only the days for which we had species identified, we’ve decided to use the average of the number of bird species identified in each one-minute recording by day, and did the same to calculate Acoustic Indices values. To test the correlation among the number of species identified per minute and acoustic indices, we performed Spearman’s correlation in two different subsets: using data from all biomes and separated by biome. We’ve calculated similarity between points based on species composition and results of acoustic indices to understand if the combination of acoustic indices can predict the species composition. For species composition we used the Jaccard index followed by a hierarchical cluster analysis. For the Acoustic Indices, we first ran a kmeans clustering (k=40), followed by 106 a calculation of a matrix of distance and then a hierarchical clustering. We used the packages ‘vegan’ (Oksanen et al., 2019) and ‘cluster’ (Maechler et al., 2015) on R statistical program (R Core Team, 2018). 3. Results 3.1. Species richness We identified bird species based on listening and viewing spectrograms, which lead us to a total of 132 bird species in the nine sampling locations. The Atlantic Forest was the biome with most species, 69, while in Caatinga we identified 53 bird species. Cerrado was the biome with less species identified, 25 (Supplementary Material − Table S1). The average number of species identified per file was also higher in Atlantic Forest (4.42 species), followed by Caatinga (2.8 species) and Cerrado (2.7 species). We also investigated the occurrence of species, and the ten most common species in each biome are presented in Table 3. Cerrado and Atlantic Forest showed few species that were registered in more than 44% of the files, while in Caatinga we observed a more equal distribution of species among files (all less than 30%). The spectral characteristics of birds’ calls was also investigated (Table 3, Figure S2), and the frequency bands used by the species revealed that in Atlantic Forest most species use frequency bands from 3-6 kHz, but few calls in higher frequency bands. In Cerrado, we observed a similar pattern, but with more species using higher frequency bands. Caatinga species presented another pattern, with a more homogeneous occupation of frequency bands. 107 Table 3 – Fist ten most common species identified in each biome, the respective percentage of files in which they appear and their call frequency range. Biomes Species % of files Frequency range Caatinga Tolmomyias flaviventris 28.7 5-12 kHz Nystalus maculatus 20.7 1-2 kHz Troglodytes musculus 19.0 1-8 kHz Eupsittula cactorum 16.7 1-12 kHz Columbina picui 13.3 0.3-0.8 kHz Todirostrum cinereum 12.0 3-12 kHz Myiarchus tyrannulus 10.7 1-5 kHz Myrmorchilus strigilatus 10.7 2-5 kHz Columbina minuta 9.3 0.3-0.8 kHz Coereba flaveola 8.7 4-11 kHz Cerrado Sicalis citrina 44.0 2-8 kHz Rhynchotus rufescens 43.0 1-3 kHz Ammodramus hummeralis 36.7 2-10 kHz Anthus nattereri 35.7 2-8 kHz Nothura maculosa 20.3 1-3 kHz Anumbius annumbi 11.7 2-5 kHz Synallaxis albescens 3.3 1-11 kHz Cariama cristata 2.3 0.5-5 kHz Embernagra platensis 2.0 2-10 kHz Neothraupis fasciata 1.6 5-12 kHz Atlantic Forest Vireo chivi 51.0 2-5 kHz Myiothlypis leocublephara 42.3 3-10 kHz Mionectes rufiventris 42.0 0.5-8 kHz Basileuterus culicivorus 37.0 3-10 kHz Conopophaga lineata 26.0 2-4 kHz Dysithamnus mentalis 20.7 0.5-4 kHz Lathrotriccus euleri 20.3 1-5 kHz Corythopis delalandi 19.3 2-4 kHz Pachyramphus polychopterus 10.7 1-8 kHz Thamnophilus caerulescens 9.3 1-3 kHz 3.2. Acoustic Indices predicting bird species richness The distribution of the acoustic indices showed different patterns among the biomes and sampling points (Figure 2). Acoustic Complexity Index presented higher mean values and more evenly distributed in the Atlantic Forest. While in ACI and BIO results, Caatinga 108 presented lower values than Atlantic Forest, and similar to Cerrado. Total Entropy (H) is higher in Caatinga, especially in point “a”, which also appear to increase the values of AR for this point. We found no overlap among sampling points (labelled as “g”, “h” and “i” in Fig. 1) considering AR values in the Atlantic Forest. Significant differences were found for all indices among biomes when sample sites were pooled together (Supplementary Material, Table S3). 109 Figure 2 – Distribution of acoustic indices values among biomes and separated into sampling points. Y axis represents normalized values of Acoustic Complexity (ACI), Acoustic Richness (AR), Total Entropy (H), Bioacoustic Index (BIO), Acoustic Diversity (ADI), and Normalized Difference Soundscape Index (NDSI), averaged by daily values. Letters “a” to “i” identify the sampling points. Significant differences were found for all indices among biomes when sample sites were pooled together (CA, CE and AF). The correlation between number of species and acoustic indices was significant for NDSI when we considered all biomes pooled together, and also within Cerrado (Table 4 and Figure 3). The Atlantic Forest showed a negative correlation between AR and species richness (rho = -0.42). In the Caatinga no index was correlated to richness. Table 4 - Spearman rho correlation coefficients between daily average of number of species and acoustic indices. Results are shown for all biomes combined and for each three biomes (Caatinga, Cerrado and Atlantic Forest). *p < 0.05 after a multiple comparison test (False Discovery Rate). NDSI ACI ADI BIO H AR All Biomes 0.31* 0.08 0.15 0.01 -0.12 -0.11 Caatinga 0.1 0.21 0.05 -0.06 0.13 0.004 Cerrado 0.35* -0.16 0.18 -0.22 0.23 -0.25 Atlantic Forest -0.003 0.05 -0.01 0.05 -0.11 -0.42* 110 111 Figure 3 – Relationship between daily averaged values of acoustic indices and number of bird species identified (normalized values). Graphs on the left were made with data from all points, shaded area represent the standard error. On the right, separated by biomes: Atlantic Forest (Blue squares), Caatinga (Red dots), Cerrado (green triangles). 3.3. Acoustic Indices predicting bird species composition Our dendrogram of similarity of bird species composition revealed a separation between the biomes, with points within the same biome being grouped together (Figure 4). The dendrogram built with Acoustic Indices values presented a similar pattern (Figure 5), with only one point grouped a different biome. Looking at the distribution of acoustic indices (Figure 1) we’ve found that this point presented higher values of H, BIO, AR and ADI compared to the other sampling points in Caatinga. The mean number of species identified per recording at point “a” is even smaller than the others in Caatinga (a= 0.36. b= 0.37. c= 0.46), so the cause could be a higher predominance of wind in the recordings. Figure 4 – Dendrogram of the similarity (calculated by Jaccard Index) of the sampling points, accordingly to the bird species composition for the Caatinga, Cerrado and Atlantic Forest. 112 Figure 5 – Dendrogram of the similarity of the sampling points, accordingly to the values of six Acoustic Indices for the Caatinga, Cerrado and Atlantic Forest. 4. Discussion We identified more bird species in the Atlantic Forest, followed by Caatinga and Cerrado. Nonetheless, despite the overall number of species was lower in Cerrado, the number of species identified in each recording was similar to Caatinga (average of Cerrado= 0.39 and Caatinga= 0.40). This result suggests predominance of fewer species in Cerrado, which was corroborated by the ranking of the most common species (Table 3), and by the pattern of low diversity with high activity indicating a strong acoustic competition and species territoriality (Izaguirre et al., 2018). The frequency bands used by birds are in agreement with the theory that states vegetation plays a role in selecting the species’ calls (Morton, 1975). According to the Acoustic Adaptation Hypothesis (AAH) (Ey & Fischer, 2009), forest areas will have species with calls concentrated in lower frequency bands, as we observed in Atlantic Forest (Figure S1). In open environments, where transmission is more efficient and soundwaves are less scattered by the vegetation, species will have calls with wider frequency range, as we observed in Caatinga (Figure S1). 113 The results from correlations between acoustic indices and species richness varied when we considered the biomes individually or pooled together. NDSI was correlated to bird richness from all biomes, and in sites within the Cerrado. This corroborates with the findings of a previous study that considered only Cerrado (Machado et al., 2017). NDSI is an index that represents more the influence of anthropic sounds on the soundscape than the number of species itself. Our results indicate that we have more human-made sounds in the Caatinga area (lower values of NDSI), which is supported by the fact that the Caatinga area sampled isn’t within a legal reserve and suffers acoustic impact from roads and other human activities nearby. Acoustic Diversity Index wasn’t correlated to species richness here, although we found a tendency when all biomes were pooled together. ADI was a good predictor of bird richness in other studies (Mammides et al., 2017; Moreno-Gómez et al., 2019). This index is assumed to capture the diversity of signals, but a concentration of signals in some frequency bands may cause incongruences in ADI values, as observed in our Atlantic Forest data with a tendency to a negative correlation with richness (Figures S1 and 3). Total Entropy (H) was considered a good indicator of bird diversity by Zhao et al. (2019), but their results were dependent on the sound unit feature, which may explain why we did not find a significant correlation in our study. H is a product of Temporal Entropy (Ht) and Spectral Entropy (Hf), which separately, were significantly correlated to richness in previous studies (Moreno-Gómez et al., 2019; Towsey et al., 2014). Ht is used in the calculation of another index used in our study, the Acoustic Richness (AR). The Acoustic Richness was developed to capture information in noisy and open environments, with low signal to noise ratio and high presence of wind and rain (Depraetere et al., 2012). Here, we found a complete separation of values of three sampling points in Atlantic Forest (Figure 2), the environment in 114 which the use of the index was, in theory, less appropriate. AR was negatively correlated to the number of species in Atlantic Forest, and we found the same negative tendency within Cerrado. Looking at how AR is calculated (see Table 2), we verified that instead of several species, one or a few highly active bird species were predominant in the Cerrado and Atlantic Forest data (Table 3) and might have inflated the values of this index (Zhao et al., 2019). The characteristics of acoustic communities are probably the reason why the our results diverge from those from a French forest (Depraetere et al., 2012). Overall, indices that consider information of distribution of energy in frequency bands are more effective in predicting species richness than indices that quantify more acoustic activity, when we consider distinct environments. The later ones can be biased by the concentration of several species in fewer frequency bands and by the amplitude values of dominant species’ calls (Gasc et al., 2015; Zhao et al., 2019). Here, Caatinga was the biome with a more even distribution of calls among frequency bands and with less predominance of species. Even so, we couldn’t find a correlation between richness and any of the indices tested, probably because our recordings were obtained by the end of the rainy season, when calling activity might be diminished. For this reason, we recommend that further investigations also include other features of the communities such as territoriality of species (that can bias the indices by calling close to microphone for a long time) and the time of the year in which the recordings are made, as calling behaviour associated with mating and territoriality increases during the reproductive period (Bradbury & Vehrencamp, 1998). Our results shown that acoustic indices could also be used to differentiate acoustic communities, with only one point being grouped in the wrong biome (point “a - Caatinga” grouped with Atlantic Forest, Figure 5). AR, BIO, ADI and H indices’ values were higher in this 115 odd sampling point than in others (Figure 2), which should indicate higher animal activity, but also might be caused by other sound sources (natural abiotic and anthropogenic). Inspecting the files, we detected that the constant presence of wind could be the cause of this odd result. Wind was already pointed as a source of inaccurate acoustic indices values, as well as anthropogenic noises, both present in the area (Depraetere et al., 2012; Fairbrass et al., 2017). 5. Conclusions This was the first study carried in different tropical biomes that aimed to test the efficacy of using acoustic indices to predict species richness, and it was interesting to compare the same methodology among different environments. The premise of an acoustic index is to be able of summarize some information in a given area, and thus they are supposed to work in a wide variety of environments. Ours and all previous studies cited here found a correlation between species richness and some of the acoustic indices used, but not always the same ones. Why some indices perform better on some biomes than in others? Some of the acoustic indices assume that we have a relationship between frequency bins occupation and biodiversity, based on the Acoustic Niche Hypothesis (Krause, 1987). Other indices are based on the measurement of overall amount of acoustic energy. These assumptions may be too general since we here show great variability among and within habitats. A careful look at the characteristics of the acoustic communities can give some guidance when making such assumptions (Zhao et al., 2019). Information about the vegetation and the likely features of the local acoustic community may also aid in generalizations about which indices should be explored. 116 The assumption that acoustic indices can be used in rapid biodiversity assessments and conservation policies itself (Burivalova et al., 2019) still need further validation since the concept of biodiversity itself that considers only species richness may be faulty, and nowadays the functional and phylogenetic diversity are also been considered (Isaac et al., 2007). Integrate those concepts to sound analyses is another challenge, since the relationship between the phylogenetic distance, dissimilarity of acoustic features, and the role of the environment in temperate (Gasc et al., 2013) and tropical habitats (Trigg, 2015) is not yet generalizable enough to be the basis for ecological assumption when it comes to soundscape analyses. 6. References Alquezar, R. D., & Machado, R. B. (2015). Comparisons Between Autonomous Acoustic Recordings and Avian Point Counts in Open Woodland Savanna. The Wilson Journal of Ornithology, 127(4), 712–723. https://doi.org/10.1676/14-104.1 Blumstein, D. T., Mennill, D. J., Clemins, P., Girod, L., Yao, K., Patricelli, G., … Kirschel, A. N. G. (2011). Acoustic monitoring in terrestrial environments using microphone arrays: Applications, technological considerations and prospectus. Journal of Applied Ecology, 48(3), 758–767. https://doi.org/10.1111/j.1365-2664.2011.01993.x Boelman, N. T., Asner, G. P., Hart, P. J., & Martin, R. E. (2007). Multi-Trophic Invasion Resistance in Hawaii: Bioacoustics, Field Surveys, and Airborne Remote Sensing. Ecological Applications, 17(8), 2137–2144. Bradbury, J. W., & Vehrencamp, S. L. (1998). Principles of Animal Communication (1st ed.). Sinauer Associates Inc. Burivalova, Z., Game, E. T., & Butler, R. A. (2019). The sound of a tropical forest. Science, 363(6422), 28–29. https://doi.org/10.1126/science.aav1902 Burivalova, Z., Towsey, M., Boucher, T., Truskinger, A., Apelis, C., Roe, P., & Game, E. T. (2017). Using soundscapes to detect variable degrees of human influence on tropical forests in Papua New Guinea. Conservation Biology, 1–29. https://doi.org/10.1111/cobi.12968 Campos-Cerqueira, M., & Aide, T. M. (2016). Improving distribution data of threatened species by combining acoustic monitoring and occupancy modelling. Methods in Ecology and Evolution, 7(11), 1340–1348. https://doi.org/10.1111/2041-210X.12599 117 Darras, K., Batáry, P., Furnas, B. J., Grass, I., Mulyani, Y. A., & Tscharntke, T. (2019). Autonomous sound recording outperforms human observation for sampling birds: a systematic map and user guide. Ecological Applications, (June), e01954. https://doi.org/10.1002/eap.1954 Depraetere, M., Pavoine, S., Jiguet, F., Gasc, A., Duvail, S., & Sueur, J. (2012). Monitoring animal diversity using acoustic indices: Implementation in a temperate woodland. Ecological Indicators, 13(1), 46–54. https://doi.org/10.1016/j.ecolind.2011.05.006 Ey, E., & Fischer, J. (2009). The “acoustic adaptation hypothesis”—a review of the evidence from birds, anurans and mammals. Bioacoustics, 19(1–2), 21–48. https://doi.org/10.1080/09524622.2009.9753613 Fairbrass, A. J., Rennett, P., Williams, C., Titheridge, H., & Jones, K. E. (2017). Biases of acoustic indices measuring biodiversity in urban areas. Ecological Indicators, 83(August), 169– 177. https://doi.org/10.1016/j.ecolind.2017.07.064 Gasc, A., Pavoine, S., Lellouch, L., Grandcolas, P., & Sueur, J. (2015). Acoustic indices for biodiversity assessments: Analyses of bias based on simulated bird assemblages and recommendations for field surveys. Biological Conservation, 191(January), 306–312. https://doi.org/10.1016/j.biocon.2015.06.018 Gasc, Amandine, Sueur, J., Pavoine, S., Pellens, R., & Grandcolas, P. (2013). Biodiversity Sampling Using a Global Acoustic Approach: Contrasting Sites with Microendemics in New Caledonia. PLoS ONE, 8(5). https://doi.org/10.1371/journal.pone.0065311 Holmes, S. B., McIlwrick, K. A., & Venier, L. A. (2014). Using automated sound recording and analysis to detect bird species-at-risk in southwestern Ontario woodlands. Wildlife Society Bulletin, 38(3), 591–598. https://doi.org/10.1002/wsb.421 Isaac, N. J. B., Turvey, S. T., Collen, B., Waterman, C., & Baillie, J. E. M. (2007). Mammals on the EDGE: conservation priorities based on threat and phylogeny. PloS One, 2(3), e296. Kasten, E. P., Gage, S. H., Fox, J., & Joo, W. (2012). The remote environmental assessment laboratory’s acoustic library: An archive for studying soundscape ecology. Ecological Informatics, 12, 50–67. https://doi.org/10.1016/j.ecoinf.2012.08.001 Krause, B. (1987). The Niche Hypothesis: How animals Taught Us to Dance and Sing, 1– 6. Lellouch, L., Pavoine, S., Jiguet, F., Glotin, H., & Sueur, J. (2014). Monitoring temporal change of bird communities with dissimilarity acoustic indices. Methods in Ecology and Evolution, 5(6), 495–505. https://doi.org/10.1111/2041-210X.12178 Machado, R. B., Aguiar, L., & Jones, G. (2017). Do acoustic indices reflect the characteristics of bird communities in the savannas of Central Brazil? Landscape and Urban Planning, 162, 36–43. https://doi.org/10.1016/j.landurbplan.2017.01.014 118 Maechler, M., Rousseeuw, P., Struyf, A., Hubert, M., Hornik, K. (2015). Package “cluster.” https://doi.org/10.1055/s-0028-1128492 Mammides, C., Goodale, E., Dayananda, S. K., Kang, L., & Chen, J. (2017). Do acoustic indices correlate with bird diversity? Insights from two biodiverse regions in Yunnan Province, south China. Ecological Indicators, 82(March), 470–477. https://doi.org/10.1016/j.ecolind.2017.07.017 Marinho, P. H., Bezerra, D., Fonseca, C. R., Antongiovanni, M., & Venticinque, E. M. (2018). Mamíferos De Médio E Grande Porte Da Caatinga Do Rio Grande Do Norte, Nordeste Do Brasil. Mastozoología Neotropical, 25(2), 345–362. https://doi.org/10.31687/saremmn.18.25.2.0.15 Morellato, L. P. C., & Haddad, C. F. B. (2016). Introduction : The Brazilian Atlantic Forest. Biotropica, 32(4), 786–792. Moreno-Gómez, F. N., Bartheld, J., Silva-Escobar, A. A., Briones, R., Márquez, R., & Penna, M. (2019). Evaluating acoustic indices in the Valdivian rainforest, a biodiversity hotspot in South America. Ecological Indicators, 103(March), 1–8. https://doi.org/10.1016/j.ecolind.2019.03.024 Morton, E. S. (1975). Ecological Sources of Selection on Avian Sounds. The American Naturalist, 109(965), 17–34. https://doi.org/10.1086/282971 Myers, N., Mittermeier, R. a, Mittermeier, C. G., da Fonseca, G. a, & Kent, J. (2000). Biodiversity hotspots for conservation priorities. Nature, 403(6772), 853–858. https://doi.org/10.1038/35002501 Oksanen, J., Blanchet, F. G., Friendly, M., Kindt, R., Legendre, P., McGlinn, D., … Wagner, H. (2019). vegan: Community Ecology Package. Retrieved from https://cran.r- project.org/package=vegan Oliveira-Filho, A. T., & Ratter, J. A. (2002). Vegetation physiognomies and woody flora of the cerrado biome. The Cerrados of Brazil: Ecology and Natural History of a Neotropical Savanna, 91–120. Olmos, F., Silva, W. A. D. G. E., & Albano, C. G. (2005). Aves em oito áreas de Caatinga no Sul do Ceará e Oeste de Pernambuco, nordeste do Brasil: composição, riqueza e similaridade. Papéis Avulsos de Zoologia (São Paulo), 45(14), 179–199. https://doi.org/10.1590/S0031-10492005001400001 Pieretti, N., Farina, A., & Morri, D. (2011). A new methodology to infer the singing activity of an avian community: The Acoustic Complexity Index (ACI). Ecological Indicators, 11(3), 868–873. https://doi.org/10.1016/j.ecolind.2010.11.005 119 Queirolo, D., & Motta-Junior, J. C. (2007). Prey availability and diet of maned wolf in Serra da Canastra National Park, southeastern Brazil. Acta Theriologica, 52(4), 391–402. https://doi.org/10.1007/BF03194237 R Core Team. (2018). R: A language and environment for statistical computing. Vienna, Austria.: R Foundation for Statistical Computing. Retamosa Izaguirre, M. I., Ramírez-Alán, O., & De la O Castro, J. (2018). Acoustic indices applied to biodiversity monitoring in a Costa Rica dry tropical forest. Journal of Ecoacoustics, 2, TNW2NP. https://doi.org/10.22261/jea.tnw2np Shonfield, J., & Bayne, E. M. (2017). Autonomous recording units in avian ecological research: current use and future applications. Avian Conservation and Ecology, 12(1), art14. https://doi.org/10.5751/ACE-00974-120114 Strassburg, B. B. N., Brooks, T., Feltran-Barbieri, R., Iribarrem, A., Crouzeilles, R., Loyola, R., … Balmford, A. (2017). Moment of truth for the Cerrado hotspot. Nature Ecology and Evolution, 1(4). https://doi.org/10.1038/s41559-017-0099 Sueur, J., Farina, A., Gasc, A., Pieretti, N., & Pavoine, S. (2014). Acoustic indices for biodiversity assessment and landscape investigation. Acta Acustica United with Acustica, 100(4), 772–781. https://doi.org/10.3813/AAA.918757 Sueur, Jérôme, Pavoine, S., Hamerlynck, O., & Duvail, S. (2008). Rapid acoustic survey for biodiversity appraisal. PLoS ONE, 3(12). https://doi.org/10.1371/journal.pone.0004065 Sugai, L. S. M., Silva, T. S. F., Ribeiro, J. W., & Llusia, D. (2019). Terrestrial Passive Acoustic Monitoring: Review and Perspectives. BioScience, 69(1), 5–11. https://doi.org/10.1093/biosci/biy147 Towsey, M., Wimmer, J., Williamson, I., & Roe, P. (2014). The use of acoustic indices to determine avian species richness in audio-recordings of the environment. Ecological Informatics, 21(July 2015), 110–119. https://doi.org/10.1016/j.ecoinf.2013.11.007 Towsey, M., Zhang, L., Cottman-Fields, M., Wimmer, J., Zhang, J., & Roe, P. (2014). Visualization of long-duration acoustic recordings of the environment. Procedia Computer Science, 29, 703–712. https://doi.org/10.1016/j.procs.2014.05.063 Towsey, M., Znidersic, E., Broken-Brow, J., Indraswari, K., Watson, D. M., Phillips, Y., … Roe, P. (2018). Long-duration, false-colour spectrograms for detecting species in large audio data-sets. Journal of Ecoacoustics, 2, IUSWUI. https://doi.org/10.22261/JEA.IUSWUI Trigg, L. (2015). Assessment of acoustic indices for monitoring phylogenetic and temporal patterns of biodiversity in tropical forests, (September). Retrieved from http://www.safeproject.net/output/assessment-of-acoustic-indices-for-monitoring- phylogenetic-and-temporal-patterns-of-biodiversity-in-tropical-forests/ 120 Villanueva-Rivera, L. J., & Pijanowski, B. C. (2016). Package “soundecology.” Http://Ljvillanueva.Github.Io/Soundecology/, CRAN, 14. Retrieved from http://ljvillanueva.github.io/soundecology/ Villanueva-Rivera, L. J., Pijanowski, B. C., Doucette, J., & Pekin, B. (2011). A primer of acoustic analysis for landscape ecologists. Landscape Ecology, 26(9), 1233–1246. https://doi.org/10.1007/s10980-011-9636-9 Wimmer, J., Towsey, M., Roe, P., & Williamson, I. (2013). Sampling environmental acoustic recordings to determine bird species richness Sampling environmental acoustic recordings to determine bird species richness, 23(September), 1419–1428. https://doi.org/10.1890/12-2088.1 Zhang, L., Towsey, M., Zhang, J., & Roe, P. (2016). Classifying and ranking audio clips to support bird species richness surveys. Ecological Informatics, 34, 108–116. https://doi.org/10.1016/j.ecoinf.2016.05.005 Zhao, Z., Xu, Z. yong, Bellisario, K., Zeng, R. wen, Li, N., Zhou, W. yang, & Pijanowski, B. C. (2019). How well do acoustic indices measure biodiversity? Computational experiments to determine effect of sound unit shape, vocalization intensity, and frequency of vocalization occurrence on performance of acoustic indices. Ecological Indicators, 107(July), 105588. https://doi.org/10.1016/j.ecolind.2019.105588 121 Supplementary Material Table S1 – Bird species identified in three biomes, CA=Caatinga, CE=Cerrado, AF=Atlantic Forest. Species Common name CA CE AF Tinamidae Crypturellus obsoletus Brown Tinamou X Crypturellus tataupa Tataupa Tinamou X Rhynchotus rufescens Red-winged Tinamou X Nothura maculosa Spotted Nothura X Accipitridae Elanus leucurus White-tailed Kite X Rupornis magnirostris Roadside Hawk X X Columbidae Columbina squammata Scaled Dove X Patagioenas picazuro Picazuro Pigeon X Patagioenas plumbea Plumbeous Pigeon X Leptotila verreauxi White-tipped Dove X X Leptotila rufaxilla Gray-fronted Dove X Strigidae Glaucidium brasilianum Ferruginous Pygmy-Owl X Apodidae Panyptila cayennensis Lesser Swallow-tailed Swift X Trogonidae Trogon surrucura Surucua Trogon X Bucconidae Nystalus maculatus Spot-backed Puffbird X Picidae Picumnus temminckii Ochre-collared Piculet X Veniliornis passerinus Little Woodpecker X Cariamidae Cariama cristata Red-legged Seriema X X Falconidae Herpetotheres cachinnans Laughing Falcon X Falco femoralis Aplomado Falcon X Psittacidae Eupsittula aurea Peach-fronted Parakeet X Eupsittula cactorum Cactus Parakeet X Forpus xanthopterygius Blue-winged Parrotlet X Brotogeris tirica Plain Parakeet X Pionus maximiliani Scaly-headed Parrot X Triclaria malachitacea Blue-bellied Parrot X Thamnophilidae Myrmorchilus strigilatus Stripe-backed Antbird X Dysithamnus mentalis Plain Antvireo X 122 Herpsilochmus rufimarginatus Rufous-winged Antwren X Thamnophilus caerulescens Variable Antshrike X Hypoedaleus guttatus Spot-backed Antshrike X Batara cinerea Giant Antshrike X Pyriglena leucoptera White-shouldered Fire-eye X Drymophila ochropyga Ochre-rumped Antbird X Conopophagidae Conopophaga lineata Rufous Gnateater X Grallariidae Grallaria varia Variegated Antpitta X Formicariidae Chamaeza campanisona Short-tailed Antthrush X Scleruridae Sclerurus scansor Rufous-breasted Leaftosser X Dendrocolaptidae Lepidocolaptes angustirostris Narrow-billed Woodcreeper X Xenopidae Xenops rutilans Streaked Xenops X Furnariidae Furnarius rufus Rufous Hornero X Lochmias nematura Sharp-tailed Streamcreeper X Automolus leucophthalmus White-eyed Foliage-gleaner X Anabazenops fuscus White-collared Foliage-gleaner X Anumbius annumbi Firewood-Gatherer X Synallaxis albescens Pale-breasted Spinetail X Cranioleuca pallida Pallid Spinetail X Pipridae Chiroxiphia caudata Swallow-tailed Manakin X Tityridae Pachyramphus polychopterus White-winged Becard X Pachyramphus validus Crested Becard X Platyrinchidae Platyrinchus mystaceus White-throated Spadebill X Rhynchocyclidae Mionectes rufiventris Gray-hooded Flycatcher X Leptopogon amaurocephalus Sepia-capped Flycatcher X Corythopis delalandi Southern Antpipit X Tolmomyias sulphurescens Yellow-olive Flycatcher X Tolmomyias flaviventris Yellow-breasted Flycatcher X Todirostrum cinereum Common Tody-Flycatcher X Poecilotriccus plumbeiceps Ochre-faced Tody-Flycatcher X Hemitriccus margaritaceiventer Pearly-vented Tody-tyrant X Tyrannidae Euscarthmus meloryphus Tawny-crowned Pygmy-Tyrant X 123 Phaeomyias murina Mouse-colored Tyrannulet X Pitangus sulphuratus Great Kiskadee X Myiodynastes maculatus Streaked Flycatcher X Megarynchus pitangua Boat-billed Flycatcher X Tyrannus melancholicus Tropical Kingbird X Empidonomus varius Variegated Flycatcher X Sublegatus modestus Southern Scrub-Flycatcher X Lathrotriccus euleri Euler's Flycatcher X Vireonidae Cyclarhis gujanensis Rufous-browed Peppershrike X Vireo chivi Chivi Vireo X Corvidae Cyanocorax cyanopogon White-naped Jay X Hirundinidae Stelgidopteryx ruficollis Southern Rough-winged Swallow X Troglodytidae Troglodytes musculus Southern House Wren X Cantorchilus longirostris Long-billed Wren X Polioptilidae Polioptila plumbea Tropical Gnatcatcher X Turdidae Turdus leucomelas Pale-breasted Thrush X Turdus rufiventris Rufous-bellied Thrush X Turdus albicollis White-necked Thrush X Motacillidae Anthus nattereri Ochre-breasted Pipit X Passerellidae Zonotrichia capensis Rufous-collared Sparrow X Ammodramus humeralis Grassland Sparrow X Parulidae Setophaga pitiayumi Tropical Parula X Basileuterus culicivorus Golden-crowned Warbler X Myiothlypis flaveola Flavescent Warbler X Icteridae Pseudoleistes guirahuro Yellow-rumped Marshbird X Thraupidae Neothraupis fasciata White-banded Tanager X Schistochlamys ruficapillus Cinnamon Tanager X Tangara desmaresti Brassy-breasted Tanager X Tangara cayana Burnished-buff Tanager X Sicalis citrina Stripe-tailed Yellow-Finch X Sicalis luteola Grassland Yellow-Finch X Coryphospingus pileatus Pileated Finch X Dacnis cayana Blue Dacnis X Coereba flaveola Bananaquit X X Embernagra platensis Great Pampa-Finch X 124 Emberizoides herbicola Wedge-tailed Grass-Finch X Saltator similis Green-winged Saltator X Saltator fuliginosus Black-throated Grosbeak X Cardinalidae Habia rubica Red-crowned Ant-Tanager X Cyanoloxia brissonii Ultramarine Grosbeak X Fringillidae Spinus magellanicus Hooded Siskin X Euphonia chlorotica Purple-throated Euphonia X Table S2 – Descriptive statistics for seven Acoustic Indices. Data is presented as mean (standard deviation). Caatinga Cerrado Atlantic Forest NDSI 0.439 (sd= 0.234) 0.604 (sd= 0.304) 0.714 (sd= 0.225) ACI 0.242 (sd= 0.198) 0.307 (sd= 0.227) 0.392 (sd= 0.292) ADI 0.425 (sd= 0.409) 0.222 (sd= 0.316) 0.537 (sd= 0.280) AEI 0.688 (sd= 0.370) 0.875 (sd= 0.265) 0.653 (sd= 0.274) BIO 0.357 (sd= 0.235) 0.263 (sd= 0.222) 0.429 (sd= 0.235) H 0.600 (sd= 0.254) 0.317 (sd= 0.298) 0.272 (sd= 0.247) AR 0.470 (sd= 0.312) 0.335 (sd= 0.308) 0.445 (sd= 0.169) Table S3 – Results of Kruskal-Wallis and Wilcox pairwise comparison for each Acoustic Indices, between the three biomes. Chi-Squared df p-value Wilcox Test NDSI 1762.7 2 <0.01 <0.01 for all pair comparison ACI 448.5 2 <0.01 <0.01 for all pair comparison ADI 1624.7 2 <0.01 <0.01 for all pair comparison AEI 2026.9 2 <0.01 <0.01 for all pair comparison BIO 1055.1 2 <0.01 <0.01 for all pair comparison H 2696.3 2 <0.01 <0.01 for all pair comparison AR 673.4 2 <0.01 <0.01 for all pair comparison 125 Figure S1 – Frequency bands used by the ten most commonly identified bird species in each biome. 126 CONCLUSÕES FINAIS Algumas das questões que a ecologia acústica busca responder são ligadas às métricas usadas como um proxy da biodiversidade de uma determinada área, da sua saúde ambiental e de como alterações decorrentes de processos temporais, espaciais, ou de mudanças no habitat podem ser identificados nas gravações de áudio. Essas questões foram abordadas pelos três capítulos desta tese. No primeiro capítulo, utilizamos múltiplos índices acústicos descritivos para mapear a distribuição temporal de sons provenientes de biofonia e geofonia, diferenciando os dois grupos animais que compuseram a sinfonia da Caatinga, aves e insetos. Outros elementos mais esporádicos nas gravacoes, como anuros, tiros, carros e animais domésticos, foram identificados em inspeções manuais, mas não constituíram fontes significativas de energia acústica a ponto de serem identificados pela metodologia empregada. A Caatinga é um ambiente que sofre forte pressão antrópica. Portanto, fontes de emissão sonora associadas a ocupação da terra, como espécies invasoras e caça, podem ser objeto de estudos futuros que se utilizem de detectores automáticos, para investigar especificamente estas fontes de som e quantificar sua presença no ambiente. O Capítulo 2 tratou de uma das alterações humanas na paisagem acústica da Caatinga. Parques eólicos são uma fonte de mudança na paisagem caatingueira crescente nos últimos anos, e a influência acústica desta atividade foi investigada. Encontramos uma diminuição da atividade de aves e insetos ligada à proximidade com os aerogeradores, fontes de emissão de ruído e que representam um disturbio tambem fisico na paisagem natural deste bioma. Espécies com vocalizações de frequências mais baixas que não estão sendo detectadas perto 127 das turbinas possivelmente tem sua comunicação comprometida ao redor das turbinas. Esse fenômeno de mudança da composição, abandono de área, ou silenciamento de espécies ao redor das turbinas deve ser compreendido em maior escala por estudos futuros. Em regiões com alta densidade de parques eólicos, onde a influência de uma linha de aerogeradores pode se sobrepor ao de uma linha vizinha, os efeitos sobre a fauna podem ser sinérgicos e mais ainda mais fortes. Nos dois primeiros capítulos, os índices acústicos foram usados com sucesso para diferenciar fontes de emissão sonora e quantificar atividade acústica, mas uma das principais premissas dessas métricas é a de ser capaz de serem usadas como um proxy de riqueza de espécies. Assim, utilizamos dados de Caatinga, Mata Atlântica e Cerrado para tentar entender, no capítulo 3, de que forma os índices se relacionam com o número de espécies de aves presentes nas áreas. Os resultados de diferentes índices variaram entre as áreas e também para todas as áreas consideradas juntas. Características da comunidade de aves, como composição de espécies e bandas de frequências mais utilizadas são fatores que podem explicar essa variação. Características da vegetação das áreas, que agem como um filtro evolutivo e sonoro para as faixas de frequência utilizadas pelas aves, podem ser um indicativo de quais índices podem ser mais adequados em diferentes ambientes. Na Caatinga, principal bioma explorado nessa tese, observamos uma limitação no uso de índices acústicos como proxy de riqueza de espécies (Cap 3). O emprego de múltiplas métricas (Caps 1 e 2), por outro lado, foi capaz de caracterizar padrões temporais, mensurar atividade e construir visualizações da comunidade sonora neste bioma. No entanto, isso demonstra uma fragilidade na generalização das metodologias de estudo de paisagens acústicas e como é importante que a validação e inspeção manuais sejam feitas. Por ter forte 128 contato com as ciências da computação, quanto maiores são os bancos de dados analisados, maior o distanciamento da realidade do ambiente. Assim, o trabalho do ecólogo é fundamental na interpretação desses resultados e na validação e escolha dos métodos a serem empregados. Essa tese deu um passo à frente nessa etapa em que se encontra a área de ecologia acústica, de indicar padrões gerais que podem ser incorporados desde a escolha das análises. No nosso caso, embora a Caatinga seja um ambiente rico em diversidade, as características do ambiente como vegetação aberta e consequente presença forte de vento são um limitante em estudos de paisagem sonora. A sazonalidade marcada e a redução da atividade dos animais devido ao clima também reduz as chances de detecções das espécies com o uso de monitoramento acústico passivo em boa parte do ano. Esse conhecimento da dinâmica das paisagens acústicas pode ser utilizado em outras áreas que apresentem características ambientais semelhantes, auxiliando na escolha de métodos na coleta e análise de dados. Por ser uma nova área de estudo, que acumula menos de uma década de contribuições substanciais à ciência e que é fortemente influenciada pelos avanços da área tecnológica, nesses quatro últimos anos vimos um aumento considerável de metodologias sendo criadas e novas perguntas a serem respondidas. Novas regiões são estudadas e, da mesma forma que a vegetação e comunidades animais, as paisagens acústicas possuem peculiaridades que ainda estão sendo exploradas. Boa parte das métricas foram desenvolvidas em regiões temperadas, onde a diversidade de animais e seus sons não é tão rica quanto nos trópicos. Esse trabalho também cobre um vazio de conhecimento da área de ecologia acústica em regiões tropicais, demonstra algumas limitações dos métodos e ainda deixa muitas perguntas em aberto para serem desenvolvidas em trabalhos futuros. 129 APÊNDICES 1. Ferreira, L. M., Oliveira, E. G., Lopes, L. C., Brito, M. R., Baumgarten, J., Rodrigues, F. H., & Sousa-Lima, R. S. (2018). What do insects, anurans, birds, and mammals have to say about soundscape indices in a tropical savanna. Journal of Ecoacoustics, 2(March), PVH6YZ. https://doi.org/10.22261/JEA.PVH6YZ 2. Oliveira, E. G. O som da natureza (2019). Ciência Hoje Crianças, ed 298, 7:9. 3. Brum, H., J. V. Campos-Silva, E. G. Oliveira (2020). Brazil oil spill response: Government inaction. Science letters, vol. 367, issue 6474. 130 View publication stats O som da natureza Concentre-se para responder: você ouve o barulho das águas de córregos e cachoeiras, das folhas das árvores quando o vento bate, das aves cantando e de outros animais que se escondem na vegetação – onde estamos? Na floresta, claro! Agora você ouve barulhos de sirenes, buzinas, máquinas furando o chão, vozes de muita gente ao mesmo tempo – onde estamos? Na cidade? Acertou! A noção de saber onde estamos somente por meio dos sons é criada por um dos nossos sentidos: a audição. O som tem importantes funções não apenas para nós, humanos, mas também para os outros animais, que se comunicam por meio dos sons que emitem. Portanto, ouvir é importante para bicho e para gente! Nós, humanos, criamos abelhas... –, o que eles querem sons para nos comunicar: dizer com os sons que emitem? pronunciamos palavras, Essa é exatamente a pergunta frases, conversamos, cantamos, que os biólogos que trabalham e fazemos isso em um ou na área de bioacústica tentam mais idiomas até! Para alguns responder. Nem sempre é uma pesquisadores, essa linguagem tarefa fácil! tão elaborada, isto é, essa capacidade tão especial de falar O que é bioacústica? (emitir som), ouvir e responder foi o que fez a nossa sociedade A bioacústica estuda os sons evoluir tanto. emitidos pelos animais para Mas, e quando falamos de se comunicar. Esses sons outros animais – como pássaros, podem ser muito variados e macacos, baleias, sapos, com as mais diversas funções, CHC ABRIL 2019 | 7 Ilustrações Jaca dependendo do animal. Aves, lado, os elefantes emitem um origens se combinam para anfíbios e insetos, por exemplo, outro tipo de som, o infrassom, formar a identidade acústica emitem sons para encontrar que é muito grave mesmo, mais de uma área. seus parceiros e defender seus grave do que o mais greve dos Complicado? Que nada! territórios. Macacos podem sons que podemos ouvir e, por Presta atenção: a identidade gritar para se comunicar dentro isso, não conseguimos escutar acústica de um lugar é como do grupo, informar onde podem também. a impressão digital de uma encontrar água ou comida. Cada pessoa. Assim como cada espécie tem a capacidade de Paisagens que falam pessoa tem uma impressão ouvir melhor o que os membros digital que não é igual a de de sua própria espécie estão Além de utilizar o som para ninguém, cada local tem sua dizendo, como se estivem estudar a comunicação entre identidade acústica própria! Na sintonizados. os animais, recentemente identidade acústica, além dos Mas sabia que os sons surgiu uma área de estudo sons dos animais, são incluídos emitidos por determinados chamada ecologia de também outros barulhos da animais nem sempre são paisagens acústicas. Opa, natureza, como os de trovões, percebidos pela audição paisagem? Como assim? ventos, chuvas... humana? Morcegos e golfinhos, Vamos relembrar o começo Como os sons da natureza por exemplo, se comunicam desse texto: a gente consegue variam de acordo com a época de uma forma que permite reconhecer uma área apenas do ano, um mesmo ambiente um localizar o outro – é o que ouvindo os sons ao redor, não pode ter sons diferentes ao chamamos ecolocalização. consegue? Pois é, quando longo do ano, certo? Assim, os A ecolocalização não é estudam paisagens acústicas, cientistas se perguntam: que percebida por nós, seres os pesquisadores não analisam animais fazem mais barulho humanos, porque esses cada som dos animais na primavera? E nas outras ultrassons são mais agudos separadamente, eles estão estações? Essas são perguntas do que o som mais agudo que mais interessados em entender que a ecologia de paisagens conseguimos escutar. Por outro como os sons de diferentes acústicas tenta responder. 8 | CHC.ORG.BR 1, 2, 3, gravando! Com essas gravações, é de formas mais rápidas possível verificar como os para entender o que está Agora, como é que os cientistas sons se modificam ao longo acontecendo nas gravações. tentam encontrar respostas? Se do tempo, comparar sons de É por isso que costumam usar você pensou que eles gravam diferentes áreas, e até mesmo programas de computador sons de diferentes locais em ter uma rápida ideia de que que transformam o som em diferentes períodos, acertou! área possui mais espécies de imagem! Mas ninguém precisa ficar animais. Hora de gritar “UAU!” e segurando um gravador no perguntar: por que transformar meio de uma floresta, por Som que vira imagem sons em imagens? Ora, porque é exemplo, por meses ou até muito mais rápido olhar para as anos. Já existem gravadores Mas quem é que ouve todas gravações, do que escutá-las. automáticos, que podem ser essas gravações? Bem, por mais deixados no local que se deseja que seja interessante ouvir os estudar por longos períodos de sons da natureza para estudá- Eliziane Garcia de Oliveira, tempo. los, os cientistas precisam Programa de Ecologia, Universidade Federal do Rio Grande do Norte. Som natural Muitos cientistas estão disponibilizando essas gravações para que todo mundo possa escutar e aprender mais sobre os sons da natureza. Quer ouvir? https://www.wikiaves.com.br/ http://www.legrandorchestredesanimaux.com/en CHC ABRIL 2019 | 9 Fotos Wikipédia Downloaded from http://science.sciencemag.org/ on January 9, 2020 Downloaded from http://science.sciencemag.org/ on January 9, 2020